Fragmentation for measuring methylation and disease

Abstract

Fragmentation of free DNA molecules is measured and used to achieve various objectives, including determining methylation at, for example, a specific site of a DNA molecule, at a specific genomic site in a reference genome of a free DNA biological sample (e.g., plasma, serum, urine, saliva) of a subject, or in a specific region in a reference genome of a biological sample (also referred to as a sample for short). Various types of fragmentation measurements may be used, such as end motifs and cleavage patterns. Another object is to determine the proportional concentration of DNA (e.g., clinically relevant DNA) of a particular tissue type. Another object is to determine a lesion in a subject using a biological sample comprising free DNA. The free DNA may be from a pathogen (e.g., a virus) in the subject or a sample of the subject. Specific tissue types of sites / regions of hypermethylation, hypomethylation, 5hmC enrichment, and 5hmC deletion may be used.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to, and is a non-provisional of, the following provisional applications: U.S. Provisional Application No. 63 / 307,622, filed on February 7, 2022, entitled “Fragmentation For Measuring Methylation And Disease”; U.S. Provisional Application No. 63 / 328,710, filed on April 7, 2022, entitled “Fragmentation For Measuring Methylation And Disease”; and U.S. Provisional Application No. 63 / 400,244, filed on August 23, 2022, entitled “Fragmentation For Measuring Methylation And Disease”, the entire contents of which are incorporated herein by reference for all purposes. Background Art

[0002] Free DNA (cfDNA) molecules are non-randomly fragmented. The association between cfDNA fragmentation patterns and nucleosome structure has been demonstrated in multiple studies (Sun et al. Proc Natl Acad Sci UAS. 2018; 115: E5106; Snyder et al. Cell. 2016; 164: 57-68). For example, the characteristic size pattern of cfDNA shows a modal frequency of about 166 bp, with smaller molecules forming a series of peaks with a periodicity of 10 bp (Lo et al. Sci Transl Med. 2010; 2: 61ra91). The size pattern of such plasma DNA fragments indicates that there is internucleosomal and intranucleosomal cutting during the release of DNA molecules into the blood circulation after cell death (e.g., via apoptosis) or other mechanisms. In addition, our group has previously reported that a subset of genomic locations were found to be preferentially cleaved during the generation of plasma DNA molecules (Chan et al. Proc Natl Acad Sci USA. 2016; 113: E8159-E8168; Jiang et al. Proc Natl Acad Sci USA. 2018; 115: E10925-E10933), and such preferential cleavage may reflect the tissue of origin of cfDNA (Jiang et al. Proc Natl Acad Sci USA. 2018; 115: E10925-E10933; Sun et al. Proc Natl Acad Sci USA. 2018; 115: E5106-E5114).

[0003] Currently, widely used strategies for methylation analysis are based on bisulfite or enzymatic conversion followed by sequencing. Such conversions involve additional DNA purification steps or DNA degradation under extreme conditions, resulting in loss of information in the DNA molecule. Summary of the Invention

[0004] Provide various embodiments for various purposes using the fragmentation of free DNA molecules. An example purpose is to determine methylation, such as at a specific site of a DNA molecule, at a specific genomic site in a reference genome of a biological sample (e.g., plasma, serum, urine, saliva) of the free DNA of an object, or at a specific region of methylation in a reference genome of a biological sample (also referred to as sample). Various types of fragmentation measurements can be used to achieve this purpose, such as end motifs and cleavage patterns, which can include cleavage information at one or more positions around a site (e.g., a CpG site).

[0005] Another example purpose is to determine the proportional concentration of DNA (e.g., clinically relevant DNA) of a specific tissue type. Various types of fragmentation measurements can be used to achieve this purpose, such as terminal motifs and cleavage patterns. Certain sites can be selected for use in fragmentation measurements. For example, sites / regions of high methylation and / or low methylation of a specific tissue type can be used. As another example, sites of 5hmC enrichment or 5hmC deletion of a specific tissue type can be used. In addition, technology can use fragmentation measurements (e.g., terminal motifs and cleavage patterns) to identify regions of 5hmC enrichment and 5hmC deletion of a specific tissue.

[0006] Another example purpose is to use a biological sample including free DNA to determine the lesion of an object. The free DNA can belong to a pathogen (e.g., a virus) in an object or object sample. Various types of fragmentation measurements can be used to achieve this purpose, such as terminal motifs and cleavage patterns. Certain sites can be selected for use in fragmentation measurements. For example, sites / regions of high methylation and / or low methylation of a specific tissue type can be used, for example, to identify lesions of a specific tissue type. As another example, similar methylation can be used, but sites / regions with different cleavage patterns between objects with lesions and those without lesions. As another example, sites / regions of 5hmC enrichment or 5hmC deletion of a specific tissue type can be used, for example, to identify lesions of a specific tissue type. Machine learning models can use such tissue-specific methylation patterns (e.g., high methylation and low methylation and 5hmC enrichment and 5hmC deletion) to distinguish different types of cancer. The free DNA fragments used for fragmentation measurement can, for example, be filtered based on size. Therefore, lesions can be detected based on the amount of terminal motifs of free DNA fragments with a specific size.

[0007] These and other embodiments of the present disclosure are described in detail below.For example, other embodiments are directed to systems, devices, and computer-readable media associated with the methods described herein.

[0008] A better understanding of the nature and advantages of the embodiments of the present disclosure may be obtained by reference to the following detailed description and accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 The positions of cleavage relative to CpG sites (also referred to as CG sites) according to embodiments of the present disclosure are shown.

[0010] Figures 2A to 2B It is shown that according to an embodiment of the present disclosure, the methylation index differs depending on the distance between the 5' end and the CpG site.

[0011] Figure 3Shown are a plurality of 4-mer terminal motifs according to embodiments of the present disclosure and the percentage of methylated CpG sites covered by fragments having a specific 4-mer terminal motif.

[0012] Figure 4 An example of a method for determining NCG and CGN terminal motifs in a region of interest using droplet digital PCR according to embodiments of the present disclosure is shown.

[0013] Figure 5 An example workflow for preparing a sequencing library by selectively amplifying DNA containing NCG or CGN terminal motifs according to embodiments of the present disclosure is shown.

[0014] Figure 6 Schematic diagram showing methylation status inference based on fragmentation patterns according to embodiments of the present disclosure.

[0015] Figure 7 Definition of a cutting mode using a cutting ratio according to an embodiment of the present disclosure is shown.

[0016] Figure 8 Shown is a comparison of cleavage patterns between methylated and unmethylated CpG sites according to embodiments of the present disclosure.

[0017] Figures 9 and 10 Shown are cleavage patterns for windows associated with different sequence contexts (including CCGs) and methylation status according to embodiments of the present disclosure.

[0018] Figure 11 Schematic showing grouping of cfDNA at adjacent CpG sites.

[0019] FIG. 12A to FIG. 12B The effect of methylation patterns with multiple adjacent CpGs on the CGN / NCG motif ratio is shown.

[0020] Figure 13 Normalization of cutting patterns using normalized cutting density according to embodiments of the present disclosure is shown.

[0021] Figure 14 Cleavage patterns using cleavage densities of various methylation combinations of two CpG sites according to embodiments of the present disclosure are provided.

[0022] Figure 15 Cleavage patterns using cleavage densities of various methylation combinations of three CpG sites according to embodiments of the present disclosure are provided.

[0023] Figure 16 Shown are the cut densities for a measurement window including CCG and methylation status according to embodiments of the present disclosure.

[0024] Figure 17 Shown is a comparison of cleavage patterns between methylated and unmethylated CpG sites using cleavage density according to embodiments of the present disclosure.

[0025] Figure 18 A workflow is presented for classifying CpG sites as hypermethylated or hypomethylated using a support vector machine (SVM) according to embodiments of the present disclosure.

[0026] Figure 19 The performance of single CpG site methylation status prediction using support vector machine (SVM) based on cleavage pattern according to embodiments of the present disclosure is demonstrated.

[0027] Figure 20 The performance of single CpG site methylation status prediction using support vector machine (SVM) based on cleavage patterns from double-stranded DNA according to embodiments of the present disclosure is demonstrated.

[0028] Figures 21A to 22 B shows the performance of single CpG site methylation status prediction using support vector machine (SVM) based on cleavage patterns in different sequence contexts according to an embodiment of the present disclosure.

[0029] Figure 22 A schematic diagram showing methylation status prediction using a CNN model based on cleavage pattern and sequence context according to an embodiment of the present disclosure is shown.

[0030] Figure 23A Demonstrate the performance of single CpG site methylation status prediction using CNN based on cleavage pattern and sequence context according to embodiments of the present disclosure. Figure 23B Shown is a comparison of the sum of the cleavage ratios at positions -1 and +1 between hypermethylated and hypomethylated CpG sites for 39 urine cell-free DNA (ucfDNA) samples.

[0031] Figure 24A Demonstrate the performance of single CpG site methylation status prediction using a CNN model based on the proximal cleavage pattern excluding position 0 and sequence context. Figure 24B Demonstrate the performance of single CpG site methylation status prediction using a CNN model based on cleavage patterns and sequence context excluding positions -1, 0, and 1.

[0032] Figures 25A to 25F Box plots showing the differences in mean cleavage ratios at various positions in plasma and urine (ucfDNA) between hypermethylated and hypomethylated CpGs.

[0033] Figures 26A to 26CShown is a comparison of various techniques for determining quantitative values ​​of methylation levels (eg, density) of CpGs with different methylation levels in eight plasma DNA samples.

[0034] Figure 27 is a flow chart of a method for determining methylation at a site based on fragmentation according to an embodiment of the present disclosure.

[0035] Figure 28 Methylation indices for CGN and NCG terminal motifs according to embodiments of the present disclosure are shown.

[0036] Figure 29 Shown are the O / E ratios of CGN and NCG terminal motifs in hypermethylated and hypomethylated regions from eight healthy control samples according to embodiments of the present disclosure.

[0037] Figure 30 It is shown that according to an embodiment of the present disclosure, a binary classification or regression is determined using a 3-mer terminal motif of a TSS region.

[0038] Figures 31A to 31B An example training procedure according to an embodiment of the present disclosure is presented.

[0039] FIG. 32A to FIG. 32B and Figures 33A to 33B Binary classification results for different k-mer terminal motifs according to embodiments of the present disclosure are shown.

[0040] Figure 34A The performance of regional methylation status prediction using SVM based on regional terminal motif frequency or O / E ratio of terminal motifs according to embodiments of the present disclosure is demonstrated.

[0041] Figure 34B Demonstrates the performance of regional methylation density estimation using SVR based on regional end motif frequencies according to embodiments of the present disclosure.

[0042] Figure 35 Normalization using the expected frequency of motifs for each region (regional normalization) according to an embodiment of the present disclosure is shown.

[0043] Figure 36 Results of binary classification using an SVM model with background-normalized (region-normalized) terminal motifs according to embodiments of the present disclosure are shown.

[0044] Figure 37 Normalization of per-molecule background using expected motif frequencies in a molecule according to embodiments of the present disclosure is demonstrated.

[0045] Figure 38Results are shown for unnormalized, region-normalized, and per-molecule-normalized 3-mer motifs according to embodiments of the present disclosure.

[0046] Figure 39 Shown are the differences in normalized frequencies (region normalization) of different 2-mer terminal motifs according to embodiments of the present disclosure.

[0047] Figure 40 Shown are the differences in normalized frequencies (region normalization) of different 3-mer terminal motifs according to embodiments of the present disclosure.

[0048] Figure 41 A flow chart illustrating a method for predicting methylation in a region according to an embodiment of the present disclosure.

[0049] Figure 42 An example workflow of a machine learning model for predicting methylation status at the single-molecule level according to embodiments of the present disclosure is presented.

[0050] Figure 43A Demonstrate the performance of single-molecule methylation status analysis using CNN based on terminal motifs, sequence context, and fragment size according to embodiments of the present disclosure.

[0051] Figure 43B Demonstrates the performance of single molecule methylation status analysis based on a CNN model using molecules with CGN and NCG terminal motifs according to embodiments of the present disclosure.

[0052] Figure 44 A flow chart illustrating a method for predicting methylation of a DNA fragment according to an embodiment of the present disclosure.

[0053] Figures 45A to 45B Box plots showing the O / E ratios of CGN and NCG motifs between 5hmC-enriched and 5hmC-depleted regions.

[0054] Figures 46A to 46B Demonstrates the performance of 5hmC prediction based on 5' 3-mer terminal motifs containing CG.

[0055] Figure 47 A flow chart illustrating a method for measuring 5hmC methylation in a region in a subject's genome using cell-free DNA molecules.

[0056] Figure 48A and Figure 48B Motif frequencies of fetal-specific hypermethylated and hypomethylated regions are shown according to embodiments of the present disclosure.

[0057] Figures 49A to 49B Demonstrating that embodiments of the present disclosure are based on the presence of placenta-specific hypermethylated CpG sites ( Figure 49A ) and placenta-specific hypomethylated CpG sites ( Figure 49B )'s plasma DNA molecules and the correlation between the CGN / NCG ratio and the fetal DNA proportion.

[0058] Figures 50A to 50B Correlation between the frequency of NCG and CGN motifs based on plasma DNA molecules derived from placenta-specific hypermethylated CpG sites (50A) and placenta-specific hypomethylated CpG sites (50B) and the proportion of fetal DNA is shown according to embodiments of the present disclosure.

[0059] Figure 51 Shown is the correlation of the fetal DNA proportion inferred by the SNP-based method (x-axis) with the fetal DNA proportion predicted using the CGN / NCG ratio from placenta-specific hypermethylated and hypomethylated CpGs (y-axis), according to embodiments of the present disclosure.

[0060] Figure 52 Schematic diagram showing the CGN / NCG motif ratio analysis based on tissue-specific methylated CpGs in blood and liver.

[0061] Figures 53A to 53B Demonstrating distinct tissue-specific hypermethylation in plasma DNA from healthy controls, liver transplant cases, and pregnant cases ( Figure 53A ) and hypomethylation ( Figure 53B ) CGN / NCG motif ratio of CpG.

[0062] Figures 54A to 54B It was demonstrated that fetal-specific methylation is reflected in the cleavage pattern of cfDNA fragmentation according to embodiments of the present disclosure.

[0063] Figure 55 and Figure 56 Marker selection, data collection, and training of an SVR model for determining the proportion of fetal DNA in a cell-free sample (eg, plasma) according to embodiments of the present disclosure are demonstrated.

[0064] Figure 57 The correlation between the fetal DNA proportion inferred by the proximal cleavage pattern and the SNP-based method according to embodiments of the present disclosure is demonstrated.

[0065] Figure 58 A flow chart illustrating a method for determining the proportional concentration of clinically relevant DNA according to an embodiment of the present disclosure.

[0066] Figures 59A to 59C Figure 3 shows genome-wide methylation density between healthy controls and HCC cases according to embodiments of the present disclosure. Figure 59A ), genome-wide CGN / NCG ratio ( Figure 59B ) and the CGN / NCG ratio in the Alu region ( Figure 59C ) box plot of the differences.

[0067] Figures 60A to 60D Shown are box plots illustrating the differences in CGN and NCG motif frequencies from the whole genome and Alu regions between healthy controls and HCC cases according to embodiments of the present disclosure.

[0068] Figure 61 Shown are the Alu region CGN / NCG ratios from healthy controls, HCC, head and neck cancer (HNSCC), colorectal cancer (CRC), lung cancer (LC), and nasopharyngeal carcinoma (NPC) according to embodiments of the present disclosure.

[0069] Figure 62 Example techniques and example structures of probes for determining NCG and CGN motifs in regions of interest using droplet digital PCR (ddPCR) according to embodiments of the present disclosure are shown.

[0070] Figures 63A to 63B Shown are the top 10 CG probes and the top 10 NCG probes for potential experimental designs according to embodiments of the present disclosure.

[0071] Figure 64 Shown is an in silico analysis of CGN / NCG ratios across healthy controls and subjects with HCC, HNSCC, CRC, LC, or NPC by ddPCR analysis utilizing probes comprising the Alu-related sequences "CGCCTGT" and "TCGCTTGA," according to embodiments of the present disclosure.

[0072] Figures 65A to 65E Shown is a ROC analysis of differentiating patients with HCC, HNSCC, CRC, LC, or NPC from healthy controls using CGN / NCG ratios obtained by in silico ddPCR analysis, according to embodiments of the present disclosure.

[0073] Figure 66 Demonstrates an example method for determining NCG and CGN motifs in a region of interest using qPCR.

[0074] Figures 67A to 67C Shows examples for the design of primers and probes.

[0075] Figures 68A to 68B Results from qPCR (A) and ddPCR (B) are shown. gDNA represents mechanically fragmented gDNA samples and was used as a control in qPCR analysis.

[0076] Figures 69A to 69BShown are the CGN / NCG motif ratios in putative hypermethylated and hypomethylated CpGs in those single-molecule real-time sequencing reads collected from healthy subjects across different size ranges.

[0077] Figures 70A to 70C Box plots showing the overall CGN / NCG motif ratios of short cfDNA fragments (<=200bp; 70A) and long cfDNA fragments (>=1000bp; 70B) in healthy subjects (Healthy), HBV carriers (HBV) and HCC cases, as well as the differences in CGN / NCG motif ratios between short cfDNA fragments (<=200bp) and long cfDNA fragments (>=1000bp) (70C).

[0078] Figures 71A to 71B Demonstrate the performance of HCC detection based on CGN / NCG motif ratio and size information.

[0079] Figure 72 A flow chart illustrating a method for determining the extent of a lesion according to an embodiment of the present disclosure.

[0080] Figure 73A and Figure 73B The performance of HCC detection based on CGN / NCG motif ratios from liver-specific hypermethylated and hypomethylated regions according to embodiments of the present disclosure is demonstrated.

[0081] Figure 74A and Figure 74B The performance of HCC diagnosis using SVM based on the frequency of 3-mer terminal motifs from liver-specific hypomethylated regions according to embodiments of the present disclosure is demonstrated.

[0082] Figures 75A to 75B Demonstrated performance for HCC detection based on CG-containing 3-mer terminal motifs from HCC-specific hypermethylated and hypomethylated CpGs.

[0083] Figure 76 Shown is a ROC analysis of distinguishing patients with HCC from those without HCC using a combined data vector from HCC-specific hypermethylated and hypomethylated CpGs.

[0084] Figure 77 Example marker selection and data collection according to embodiments of the present disclosure are presented.

[0085] Figure 78 A workflow is presented that includes generating cutting patterns of multiple objects for training a machine learning model according to an embodiment of the present disclosure.

[0086] Figure 79A 、 Figure 79B and Figure 80 According to an embodiment of the present disclosure, by using a methylation site from placenta specific hypermethylation ( Figure 79A ), placenta-specific hypomethylated sites ( Figure 79B ) or a combination of placenta-specific hypermethylated and hypomethylated sites ( Figure 80 ) were analyzed by PCA on pregnant and non-pregnant subjects.

[0087] Figure 81 Demonstrate the performance of HCC detection based on cutting pattern.

[0088] Figure 82 Schematic diagram showing the definition of tissue-informative methylation markers.

[0089] Figure 83A We present an unsupervised clustering analysis of HCC and CRC patients based on 5' CG-containing 3-mer terminal motifs associated with tissue-informative and tissue-consensus methylated CpG sites. Figures 83B to 83C The performance of an HCC classifier based on a 5' CG-containing 3-mer terminal motif is shown to be associated with tissue-informative and tissue-consensus methylated CpG sites.

[0090] Figure 84 Flowchart of a method for determining the extent of disease in a specific tissue using tissue-specific differentially methylated regions according to an embodiment of the present disclosure.

[0091] Figures 85A to 85B Demonstrated performance for HCC detection based on 5' CG-containing 3-mer terminal motifs from HCC-specific CpGs enriched and depleted in 5hmC.

[0092] Figures 86A to 86B The performance of EBV-associated lymphoma detection based on CGN / NCG ratios of lymphoma-specific hypermethylated and hypomethylated CpG sites from plasma EBV DNA according to embodiments of the present disclosure is demonstrated.

[0093] Figures 87A to 87B Shown is a ROC analysis of differentiating mononucleosis (IM) patients from EBV-related lymphoma patients using CGN and NCG motif frequencies based on lymphoma-specific hypomethylated (A) and hypermethylated (B) regions, according to embodiments of the present disclosure.

[0094] Figures 88A to 88B The performance of NPC detection based on CGN / NCG ratios of NPC-specific hypermethylated and hypomethylated CpG sites in EBV DNA according to embodiments of the present disclosure is demonstrated.

[0095] Figures 89A to 89BShown are ROC analyses using the frequencies of CGN and NCG motifs from NPC-specific hypomethylated (A) and hypermethylated (B) regions to distinguish IM patients from NPC patients.

[0096] Figures 90A to 90B The distribution of cleavage ratios from NPC and non-NPC cases at the -1 and +1 positions of each CpG site where the methylation density is more than 80% in both NPC and non-NPC cases is shown.

[0097] Figure 91A Shows the distribution of the sum of the cut ratios at positions -1 and +1. Figure 91B Demonstrates a preference for cleavage in the presence of enzyme deficiency.

[0098] Figures 92A to 92D Shown are box plots illustrating the differences in EBV DNA cleavage ratios between NPC and non-NPC cases at the −1 and +1 informative positions in the training and test sets.

[0099] Figures 93A to 93B Shown are box plots illustrating the differences in EBV DNA cleavage ratios between NPC and non-NPC cases at -1 and +1 informative positions in the training set (A) and the test set (B).

[0100] Figures 94A to 94B Demonstrated diagnostic ability to distinguish NPC cases from non-NPC cases by using the cleavage ratio of EBV DNA at −1 and +1 informative positions and the proportion of EBV DNA in the training set (A) and the test set (B).

[0101] Figure 95 Demonstrate the diagnostic ability to distinguish NPC cases from non-NPC cases by using different metrics in training and testing datasets.

[0102] Figure 96 Schematic diagram showing informative CpG selection based on the adjusted CGN / NCG motif ratio in the training set and diagnosis of NPC in the test set.

[0103] Figure 97A Shown are the methylation indices of informative CpGs measured by bisulfite sequencing between non-NPC cases and NPC cases. Figure 97B Shown are the adjusted CGN / NCG motif ratios of informative CpGs in the EBV genome between non-NPC and NPC individuals (training set).

[0104] Figure 98A Shown are the adjusted CGN / NCG motif ratios of informative CpGs in the EBV genome between non-NPC and NPC individuals (test set). Figure 98BThe positive predictive values ​​achieved by PCR-based analysis, methods based on EBV DNA ratio and size ratio, and a combined EBV DNA ratio, size ratio, and cleavage pattern method are presented.

[0105] Figure 99 A flow chart illustrating a method for determining the extent of a disease in a subject using a biological sample comprising free DNA molecules from the subject and from a virus.

[0106] Figure 100A and Figure 100B It is demonstrated that according to embodiments of the present disclosure, CGN and NCG end motif frequencies can be used to predict genomic imprinting.

[0107] Figure 101 The performance of differentiating methylated from unmethylated regions depending on sequencing depth according to embodiments of the present disclosure is demonstrated.

[0108] Figure 102 Demonstrates that the relative percentage of sequenced reads that align to certain genomic regions of interest can be enriched by selectively analyzing a subset of DNA molecules with specific terminal motifs at their ends.

[0109] Figure 103 An exemplary workflow for preparing a sequencing library by selectively amplifying DNA containing NCG or CGN terminal motifs from a region of interest according to embodiments of the present disclosure is provided.

[0110] Figure 104 Shown are the CGN / NCG motif ratios of putative hypermethylated and hypomethylated CpGs in sequencing libraries of plasma DNA prepared using 4 PCR cycles, 8 PCR cycles, and no PCR, respectively.

[0111] Figure 105 Shown is a comparison of cleavage patterns between hypermethylated and hypomethylated CpG sites in urine cfDNA samples from healthy controls.

[0112] Figures 106A to 106D Shows the methylation density across the genome, Alu regions, and CpG islands in urine cfDNA of healthy controls ( Figure 106A ), CGN / NCG motif ratio ( Figure 106B ), CGN / C^G motif ratio ( Figure 106C ), CGN / (NCG+C^G) motif ratio ( Figure 106D Each dot represents a urine cfDNA sample.

[0113] Figures 108A to 108B Demonstrates a novel approach based on the presence of kidney-specific hypermethylated CpG sites ( Figure 108A ) and kidney-specific hypomethylated CpG sites ( Figure 108B ) correlation between the donor DNA ratio and the CGN / NCG motif ratio of urine cfDNA molecules.

[0114] Figure 109 Correlation between donor DNA proportion and normalized CGN / NCG motif ratio in urine cfDNA from kidney transplant patients is shown.

[0115] Figures 110A to 110B Demonstrates a novel approach based on the presence of hypermethylated CpG sites derived from blood cells ( Figure 110A ) and blood cell-specific hypomethylated CpG sites ( Figure 110B ) correlation between the donor DNA ratio and the CGN / NCG motif ratio of urine cfDNA molecules.

[0116] Figure 111 Correlation between donor DNA proportion and normalized CGN / NCG motif ratio in urine cfDNA from bone marrow transplant patients is shown.

[0117] Figures 112A to 112B Figure 2 Box plots showing the CGN / NCG motif ratios of bladder cancer-specific hypermethylated and hypomethylated CpGs in healthy controls (Control), patients with low-grade non-muscle invasive bladder cancer (NMIBC LG), high-grade non-muscle invasive bladder cancer (NMIBC HG), and muscle invasive bladder cancer (MIBC).

[0118] Figure 113 Figure 3 Box plots showing the normalized CGN / NCG motif ratios of bladder cancer-specific hypermethylated and hypomethylated CpGs in healthy controls (Control), patients with low-grade non-muscle invasive bladder cancer (NMIBC LG), high-grade non-muscle invasive bladder cancer (NMIBC HG), and muscle invasive bladder cancer (MIBC).

[0119] Figure 114 Schematic diagram of SVM model training for bladder cancer detection based on cleavage patterns from bladder cancer-specific hypermethylated and hypomethylated CpG sites.

[0120] Figures 115A to 115B Demonstrate the performance of bladder cancer detection based on cutting patterns.

[0121] Figures 116A to 116B Demonstrated performance for bladder cancer detection based on a 5' CG-containing 3-mer terminal motif.

[0122] Figure 117 Schematic representation of bladder cancer cases and cancer-free controls.

[0123] Figures 118A to 118BBladder cancer diagnosis using the number of regions with significantly low normalized CGN / NCG motif ratios (ie, motif variations) is shown.

[0124] Figure 119 Residual disease diagnosis using the amount of bins displaying significantly low normalized CGN / NCG motif ratios (motif variations) in urine cfDNA from bladder cancer patients is demonstrated.

[0125] Figure 120 Schematic diagram showing the definition of tissue-informative methylation signatures using bladder cancer as an example.

[0126] Figure 121 Schematic showing multiple cancer clusters utilizing CG-containing terminal motifs.

[0127] Figure 122 Unsupervised clustering analysis of patients with bladder, kidney, and prostate cancer based on 5' CG-containing 3-mer terminal motifs associated with tissue-informative and tissue-consensus methylated CpG sites is presented.

[0128] Figure 123 Schematic diagram showing the classification of multiple cancer types using multiple binary classifications.

[0129] Figures 124A to 124B The performance of a bladder cancer classifier based on 5' CG-containing 3-mer terminal motifs associated with tissue-informative CpG sites and tissue-consensus CpG sites is presented.

[0130] Figures 125A to 125D The performance of renal and prostate cancer classifiers based on 5' CG-containing 3-mer terminal motifs associated with tissue-informative CpG sites and tissue consensus CpG sites is shown.

[0131] Figures 126A to 126B Schematic illustration showing the classification of multiple cancers using a CNN model based on CG-containing motifs.

[0132] Figure 127 A measurement system according to an embodiment of the present invention is shown.

[0133] Figure 128 A block diagram is shown of an example computer system that may be used with systems and methods according to embodiments of the present invention. the term

[0134] A "tissue" corresponds to a group of cells grouped together as a functional unit. More than one type of cell can be found in a single tissue. Different types of tissues can be composed of different types of cells (e.g., liver cells, alveolar cells, or blood cells), but can also correspond to tissues from different organisms (mother versus fetus) or to healthy cells versus tumor cells. A "reference tissue" can correspond to a tissue used to determine tissue-specific methylation levels or tissue-specific fragmentation patterns. Multiple samples of the same tissue type from different individuals can be used to determine tissue-specific methylation levels for that tissue type.

[0135] "Biological sample" refers to any sample obtained from a subject (e.g., a human (or other animal), such as a pregnant woman, an individual with cancer or other disease, or an individual suspected of having cancer or other disease, an organ transplant recipient, or a subject suspected of having a disease process involving an organ (e.g., the heart in myocardial infarction, the brain in stroke, or the hematopoietic system in anemia)) and containing one or more nucleic acid molecules of interest. A biological sample can be a body fluid, such as blood, plasma, serum, urine, vaginal fluid, fluid from scrotal edema (e.g., testicles), vaginal washings, pleural effusion, ascites, cerebrospinal fluid, saliva, sweat, tears, sputum, bronchoalveolar lavage fluid, breast milk, aspirates from different parts of the body (e.g., thyroid, breast), intraocular fluid (e.g., humoral fluid), etc. Stool samples can also be used. In various embodiments, a majority of the DNA in a biological sample that has been enriched for free DNA (e.g., a plasma sample obtained via a centrifugation protocol) can be free, for example, more than 50%, 60%, 70%, 80%, 90%, 95% or 99% of the DNA can be free. The centrifugation protocol can include, for example, 3,000g×10 minutes to obtain a liquid portion, followed by, for example, 30,000g for another 10 minutes to remove remaining cells. As part of the analysis of the biological sample, a statistically significant number of free DNA molecules of the biological sample can be analyzed (e.g., to provide accurate measurements). In some embodiments, at least 1,000 free DNA molecules are analyzed. In other embodiments, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 or more free DNA molecules can be analyzed. At least the same number of sequence reads can be analyzed.

[0136] "Clinically relevant DNA" refers to DNA of a specific tissue origin to be measured, for example to determine the proportional concentration of such DNA or to classify the phenotype of a sample (e.g., plasma). Examples of clinically relevant DNA are fetal DNA in maternal plasma or tumor DNA in patient plasma or other samples with free DNA. Another example includes measurement of the amount of transplant-associated DNA in the plasma, serum, or urine of a transplant patient. Further examples include measurement of the proportional concentration of hematopoietic and non-hematopoietic DNA in the subject's plasma, or the proportional concentration of liver DNA fragments (or other tissues) in a sample, or the proportional concentration of brain DNA fragments in cerebrospinal fluid.

[0137] "Sequence read" refers to a series of nucleotide sequences sequenced from any part or all of a nucleic acid molecule. For example, a sequence read can be a short nucleotide sequence (e.g., 20 to 150 nucleotides) sequenced from a nucleic acid fragment, a short nucleotide sequence at one or both ends of a nucleic acid fragment, or sequencing of an entire nucleic acid fragment present in a biological sample. Sequence reads can be obtained in various ways, such as using sequencing techniques or using probes (e.g., in hybridization arrays or capture probes that can be used in microarrays) or amplification techniques (e.g., polymerase chain reaction (PCR) or linear amplification or isothermal amplification using a single primer). Example sequencing techniques include high-throughput parallel sequencing, targeted sequencing, Sanger sequencing, sequencing by ligation, ion semiconductor sequencing, and single-molecule sequencing (e.g., using nanopore sequencing, or single-molecule real-time sequencing (e.g., from Pacific Biosciences)). Example PCR techniques include real-time PCR and digital PCR (e.g., droplet digital PCR). As part of a biological sample analysis, a statistically significant number of sequence reads can be analyzed, for example, at least 1,000 sequence reads can be analyzed. As other embodiments, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 or more sequence reads can be analyzed.

[0138] A sequence read can include an "end sequence" associated with the end of a fragment. The end sequence can correspond to the outermost N bases of the fragment, for example, 1 to 30 bases from the end of the fragment. If the sequence read corresponds to the entire fragment, the sequence read can include two end sequences. When paired-end sequencing provides two sequence reads corresponding to the ends of the fragment, each sequence read can include one end sequence.

[0139] "Sequence motif" can refer to a short, recurring base pattern in a DNA fragment (e.g., a free DNA fragment). A sequence motif can appear at the end of a fragment and is therefore part of or includes a terminal sequence."Terminal motif" (also referred to as "terminal sequence motif") can refer to the motif of a terminal sequence, which may preferentially appear at the end of a DNA fragment of a particular type of tissue.Terminal motifs can also appear just before or after the end of a fragment and therefore still correspond to terminal sequences.Nuclease can have a specific cutting preference for a specific terminal motif and a second best cutting preference for a second terminal motif.The number of nucleotides (nt) at the end of the fragment for analysis can be, for example, but not limited to, 1nt, 2nt, 3nt, 4nt, 5nt, 6nt, 7nt, 8nt, 9nt, and 10nt or higher.In some embodiments, a fragment terminal motif can be defined by one or more nucleotides spanning a position near the end of the fragment.A fragment terminal motif can be defined by one or more nucleotides of the genomic locus aligned around the end of the fragment in a reference genome.

[0140] " site " (also referred to as " genomic site ") corresponds to a single site, which can be a single base position or one group of related base positions, such as a CpG site, a TSS site, a DNA enzyme hypersensitive site or a larger group of related base positions." locus " can correspond to the region comprising a plurality of sites. The locus can only include a site so that the locus is equivalent to a site in this case. Various regions, sites or loci can be analyzed, such as 50, 100, 200, 500, 1,000, 5,000, 10,000, 50,000, 100,000, 500,000, 1000,000 or larger. Various techniques can determine that the DNA molecule is positioned with reference to the position of one or more genomes in the genome, such as sequence reads and with reference to the alignment of genome or using position-specific probes. Position determination can be for some or all of the genome, such as if only analyzing a part for genome. As an example, the amount of the genome analyzed can be greater than 0.01%, 0.1%, 1%, 5%, 10%, or 50%.A "cleavage site" can refer to a location where DNA is cleaved by a nuclease, thereby generating DNA fragments.

[0141] " Cutting pattern " can refer to the amount of the fragment of two or more positions that the end is located in the window around the site (for example, CpG site).The amount of the fragment can correspond to the different categories according to the end motif (for example, position 0 and -1 are CGN and NCG respectively).The amount can be standardized, for example, using the sequencing depth at each position, the depth in the region or the number of the fragment of the end in the region.This type of standardized amount at a single position can be referred to as cutting ratio, cutting amount or cutting density.In one example, the cutting pattern can be defined as the ratio pattern between the fragment end and the sequencing depth of the genomic coordinates in the window related to the CpG site, which can be used to infer the methylation pattern of the CpG site.Various types of standardization can be used, as described herein.The window can include, but is not limited to, X nucleotides (that is, X-nt) upstream of the CpG site and Y nucleotides (that is, Y-nt) downstream.The values ​​of X and Y can be 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 30, 40, 50, 100, 1000, 5000 etc. The window can cover a range of nucleosome sizes upstream and downstream of the CpG site, for example, -160 nt to 160 nt.

[0142] The term "allele" refers to different DNA sequences at the same entity genome locus, which may or may not cause different phenotypic traits. In any specific diploid organism with two copies of each chromosome (except the sex chromosomes in male human subjects), the genotype of each gene includes the allele pair present at the locus, which is identical in homozygotes and different in heterozygotes. The colony or species of an organism typically includes multiple alleles on each locus in various individuals. The genomic locus found to exceed one allele in a colony is referred to as a polymorphic site. The allelic variation on the locus can be measured as the number (that is, the degree of polymorphism) of the alleles present in the colony or the ratio (that is, heterozygosity) of heterozygotes. As used herein, the term "polymorphism" refers to any inter-individual variation in the human genome, regardless of its frequency. The example of such variation includes, but is not limited to, single nucleotide polymorphism, simple tandem repeat polymorphism, insertion-deletion polymorphism, mutation (which can be pathogenic) and copy number variation. As used herein, the term "haploid" refers to a combination of alleles on multiple loci that are passed on together on the same chromosome or chromosome region. A haploid can refer to only a pair of loci, or to a chromosome region, or to an entire chromosome or chromosome arm.

[0143] The term "sequence context" can refer to the base composition (A, C, G or T) and base order in a section of DNA. Such a section of DNA can be around a CpG site that is subjected to a methylation analysis or as a target of a methylation analysis. For example, sequence context can refer to bases upstream and / or downstream of a CpG site subjected to a methylation analysis. For example, sequence context can include a K-polymer matrix that specifies multiple examples of each K-polymer in a test free DNA molecule, where K is an integer. When K is equal to 2, the K-polymer matrix is ​​a dinucleotide matrix specifying multiple examples of dinucleotide pairs in a test free DNA molecule. As another example, sequence context can include sequence information (i.e., complete sequence) for each nucleotide from a test free DNA molecule. Single-hot encoding can be used for the sequence to generate a 0 and 1 matrix of size 4 × N, where N is the number of bases in the sequence.

[0144] There are various types of methylation. Methylation has been found on cytosine, adenine, thymine, and guanine, including 5mC (5-methylcytosine), 4mC (N4-methylcytosine), 5hmC (5-hydroxymethylcytosine), 5fC (5-formylcytosine), 5caC (5-carboxycytosine), 1mA (N1-methyladenine), 3mA (N3-methyladenine), 7mA (N7-methyladenine), 3mC (N3-methylcytosine), 2mG (N2-methylguanine), 6mG (O6-methylguanine), 7mG (N7-methylguanine), 3mT (N3-methylthymine), and 4mT (O4-methylthymine). In vertebrate genomes, 5mC is the most common type of base methylation, followed by guanine (i.e., in a CpG context).

[0145] The "methylation index" or "methylation status" of each genomic site (e.g., a CpG site) can refer to the ratio of the number of DNA fragments that show methylation at the site (e.g., as determined by sequence readings or probes) to the total number of readings covering the site. The methylation status can refer to whether a particular site is methylated at a particular site of a DNA fragment. A "reading" can correspond to information obtained from a DNA fragment (e.g., the methylation status at a site). Readings can be obtained using reagents (e.g., primers or probes) that preferentially hybridize to DNA fragments of a particular methylation state. Typically, such reagents are applied after treatment with a method for differentially modifying or differentially recognizing DNA molecules based on the methylation status of the DNA molecule, such as bisulfite conversion, or methylation-sensitive restriction enzymes, or methylation-binding proteins, or anti-methylcytosine antibodies, or single-molecule sequencing techniques that recognize methylcytosine and hydroxymethylcytosine.

[0146] The "methylation density" of a region or a group of sites can refer to the number of reads showing methylation at sites within the region (also referred to as a bin) or group of sites divided by the total number of reads covering the sites in the region or group of sites. A region can include one or more sites of interest, including at least 1, 2, 3, 4, 5, 10, 20, 50, 100, 200, 500, and 1,000 sites. The sites can have a specified characteristic, such as being CpG sites. Thus, the "CpG methylation density" of a region can refer to the number of reads showing CpG methylation divided by the total number of reads covering the CpG sites in the region (e.g., a specific CpG site, a CpG island, or CpG sites within a larger region). For example, the methylation density per 100 kb bin in the human genome can be determined based on the ratio of the total number of unconverted cytosines (which correspond to methylated cytosines) at the CpG sites after bisulfite treatment relative to all CpG sites covered by sequence reads for the 100 kb region. This analysis can also be performed on other box sizes, such as 500bp, 5kb, 10kb, 50-kb or 1-Mb. A region can be a whole genome or a chromosome or a part of a chromosome (such as a chromosome arm). When a region only includes CpG sites, the methylation index of the CpG sites is identical to the methylation density of the region. The "ratio of methylated cytosine" can refer to the number of cytosine sites "C" that are displayed as methylation (such as after bisulfite conversion without conversion) relative to the total number of cytosine bases analyzed, i.e., including the cytosine in the region except the CpG background. The ratio of methylation index, methylation density and methylated cytosine is an example of a "methylation level". In addition to bisulfite conversion, other methods known to those skilled in the art can be used to detect the methylation status of DNA molecules, including but not limited to enzymes sensitive to methylation status (e.g., methylation-sensitive restriction enzymes), methylation binding proteins, single molecule sequencing using a platform sensitive to methylation status (e.g., nanopore sequencing (Schreiber et al. Proc Natl Acad Sci USA 2013; 110: 18910-18915) or by Pacific Biosciences single molecule real-time sequencing analysis (Tse et al. Proc Natl Acad Sci USA 2021; 118: e2019768118)).

[0147] The term "hypomethylation" can refer to a site or a group of sites (e.g., regions) that are equal to or lower than methylation levels. If the methylation level is lower than the critical value, the site in the genome can be considered as unmethylated. The term "hypermethylation" can refer to a site or a group of sites (e.g., regions) that are equal to or higher than methylation levels. If the methylation level is greater than the critical value, the site in the genome can be considered as methylated.

[0148] Differentially methylated regions (DMRs) are genomic regions with different DNA methylation states across two or more biological samples. Different DNA methylation states can be defined by specific differences in methylation index or density, such as (but not limited to) 1%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, 99%, etc.

[0149] The term "5hmC enriched" may refer to a site or a group of sites (e.g., a region) that has a relatively high proportion of 5hmC methylated molecules relative to other types of methylation. The term "5hmC depleted" may refer to a site or a group of sites (e.g., a region) that has a relatively low proportion of 5hmC methylated molecules relative to other types of methylation.

[0150] "Relative frequency" (also simply referred to as "frequency") can refer to a ratio (e.g., percentage, fraction, or concentration). In particular, the relative frequency of a particular terminal motif (e.g., CCGA or just a single base) or a group of terminal motifs (e.g., CGN or NCG) can provide the proportion of free DNA fragments in a sample that are associated with one or more of the group of terminal motifs.

[0151] An "aggregate value" can refer to, for example, a general property of the relative frequencies of a set of terminal motifs. Examples include, for example, the mean, median, sum of relative frequencies, variation between relative frequencies (e.g., entropy, standard deviation (SD), coefficient of variation (CV), interquartile range (IQR), or a percentile cutoff between different relative frequencies (e.g., 95th or 99th percentile)), or a difference (e.g., distance) from a reference pattern of relative frequencies, as can be implemented in a cluster analysis. As another example, the aggregate value can include an array / vector of relative frequencies that can be compared to a reference vector (e.g., representing a multidimensional data point).

[0152] The term "sequencing depth" refers to the number of times a locus is covered by sequence reads aligned with the locus. A locus may be as small as a nucleotide, or as large as a chromosome arm, or as large as an entire genome. Sequencing depth can be expressed as 50x, 100x, etc., where "x" refers to the number of times a locus is covered by sequence reads. Sequencing depth can also be applied to multiple loci or whole genomes, in which case x can refer to the average number of times a locus or haploid genome or whole genome is sequenced, respectively. Ultra-deep sequencing can refer to a sequencing depth of at least 100x.

[0153] "Calibration sample" can correspond to a biological sample with a known characteristic (measurement value). Example characteristics include methylation levels or the ratio concentration of clinically relevant DNA. The ratio concentration of clinically relevant DNA (e.g., tissue-specific DNA fraction) can be known or determined by a calibration method, such as using tissue-specific alleles, such as in transplantation, where alleles that exist in the donor genome but are lacking in the recipient genome can be used as markers for transplanted organs. As an example, the methylation level can be the methylation level of a site in a region, genome, or genome or on a fragment. Methylation-aware analysis can be used, such as methylation-aware sequencing or PCR to determine the methylation level. Example methylation-aware sequencing can include bisulfite sequencing or single-molecule technology, such as using nanopores. The calibration sample can have a separate measurement value (e.g., the amount of a fragment with a specific end motif or cleavage pattern), and it can be determined that the desired measurement value can be related to it.

[0154] "Calibration data points" include "calibration values" (e.g., the amount of fragments having a specific end motif or cleavage pattern) and other measured or known values ​​of the desired assay for the test sample. Calibration values ​​can be determined based on various types of data measured from the DNA molecules of the sample. Calibration values ​​correspond to parameters related to the desired properties (e.g., classification of genetic disorders or methylation levels). For example, calibration values ​​can be determined based on the measured values ​​determined for the calibration sample, for which the desired properties are known. Calibration data points can be defined in a variety of ways, such as as discrete points or as calibration functions (also referred to as calibration curves or calibration surfaces). Calibration functions can be derived from additional mathematical transformations of calibration data points.

[0155] "Separation value" corresponds to a difference or ratio involving two values, such as the difference or ratio of two ratio proportions or two methylation levels. The separation value can be a simple difference or ratio. As an example, a positive ratio of x / y and a positive ratio of x / (x+y) are separation values. The separation value can include other factors, such as a multiplication constant. As another example, a difference or ratio of a function of the following values ​​can be used, such as the difference or ratio of the natural logarithm (ln) of two values. The separation value can include a difference and a ratio. The separation value can be compared with a critical value to determine whether the separation between the two values ​​is statistically significant.

[0156] As used herein, the term "parameter" means a numerical value that characterizes a quantitative data set and / or a numerical relationship between quantitative data sets. For example, the ratio (or a function of the ratio) between a first amount of a first nucleic acid sequence and a second amount of a second nucleic acid sequence is a parameter. Parameters can be used to determine any classification described herein, such as any classification for fetal, cancer, or transplant analysis. "Separation value" (e.g., of relative frequency) and "total value" are two examples of parameters (also referred to as metrics) that provide a sample measure that varies between different classifications (states), and therefore can be used to determine different classifications. For example, when a difference is obtained between a set of relative frequencies of a sample and a set of reference relative frequencies, for example, a cluster assessment can be used, and the total value can be a separation value.

[0157] As used herein, the term "classification" refers to any number or other character associated with a particular characteristic of a sample. For example, the symbol "+" (or the word "positive") can indicate that a sample is classified as having a deletion or an amplification. Classification can be binary (e.g., positive or negative) or have more classification levels (e.g., a scale from 1 to 10 or 0 to 1).

[0158] The terms "cutoff value" and "critical value" refer to predetermined numerical values ​​used in the operation. For example, the cutoff size can refer to a size, and fragments exceeding this size will be excluded. A critical value can be a value, and a specific classification is applicable above or below this value. Any of these terms can be used in any of these situations. A cutoff value or a threshold value can be a "reference value" representing a specific classification or distinguishing two or more classifications or derived from this reference value. Such reference values ​​can be measured in various ways, as will be understood by the technician. For example, a metric can be determined for the object of two different groups with different known classifications, and a reference value can be selected as a representative (e.g., mean value) of a classification or a value (e.g., selected to obtain desired sensitivity and specificity) between two clusters of a metric. As another example, a reference value can be determined based on a statistical simulation of a sample. Specific cutoff values, critical values, reference values, etc. can be determined based on required accuracy (e.g., sensitivity and specificity).

[0159] The term "cancer grade" can refer to whether cancer exists (i.e., exists or does not exist), the stage of cancer, the size of the tumor, whether there is metastasis, the total tumor load of the body, the response of cancer to treatment, and / or other measures of cancer severity (e.g., cancer recurrence). Cancer grade can be a number or other sign, such as a symbol, letter, and color. Grade can be zero. Cancer grade can also include a condition (status) before deterioration or before cancer. Cancer grade can be used in a variety of ways. For example, screening can check whether a previously unknown person with cancer has cancer. Assessment can investigate someone who has been diagnosed with cancer to monitor the progress of cancer, the effectiveness of research therapy, or determine prognosis over time. In one embodiment, prognosis can be expressed as the probability that the patient dies from cancer, or the probability of cancer progression after a specified period or time, or the probability or degree of cancer metastasis. Detection can mean 'screening' or can mean checking whether someone with the characteristics of cancer (e.g., symptoms or other positive tests) has cancer.

[0160] "Pathology grade" can refer to the amount, extent, or severity of a lesion associated with an organism, wherein the grade can be as described above for cancer. Another example of a lesion is rejection of a transplanted organ. Other example lesions can include autoimmune attacks (e.g., lupus nephritis that damages the kidneys or multiple sclerosis that damages the central nervous system), inflammatory diseases (e.g., hepatitis), fibrotic processes (e.g., cirrhosis), fatty infiltration (e.g., fatty liver disease), degenerative processes (e.g., Alzheimer's disease), and ischemic tissue damage (e.g., myocardial infarction or stroke). The healthy state of a subject can be considered a classification of no lesion.

[0161] A "machine learning model" may refer to a software module configured to operate on one or more processors to provide a classification or numerical value of a characteristic of one or more samples. One example type of model is supervised learning that may be used with embodiments of the present disclosure. Example supervised learning models can include different methods and algorithms, including analytical learning, artificial neural networks, backpropagation, boosting (common algorithm), Bayesian statistics, case-based reasoning, decision tree learning, inductive logic programming, Gaussian process regression, genetic programming, data grouping methods, kernel estimator, learning automata, learning classification systems, minimum information length (decision trees, decision diagrams, etc.), multilinear subspace learning, Naive Bayes classifier, maximum entropy classifier, conditional random fields, nearest neighbor algorithm, probabilistic approximately correct learning (PAC) learning, ripple down rule, knowledge acquisition methods, symbolic machine learning algorithms, sub-symbolic machine learning algorithms, minimum complexity machine (MCM), random forest, classifier ensemble, ordered classification, data preprocessing, handling imbalanced data sets, statistical relational learning or Proaftn (a multi-criteria classification algorithm). Models may include linear regression, logistic regression, deep recurrent neural networks (e.g., long short-term memory, LSTM), hidden Markov models (HMM), linear discriminant analysis (LDA), k-means clustering, density-based clustering with noise (DBSCAN), random forest algorithms, support vector machines (SVM), or any model described herein. Various cost / loss functions defining the error from known labels (e.g., least squares and absolute difference from known classifications) and various optimization techniques may be used in various ways, for example, using backpropagation, steepest descent, conjugate gradient, and Newton and quasi-Newton techniques to train supervised learning models.

[0162] The term "about" or "approximately" can mean within an acceptable error range for a particular value as determined by one of ordinary skill in the art, which error range will depend in part on the manner in which the value is measured or determined, i.e., the limitations of the measurement system. For example, according to practice in the art, "about" can mean within 1 or more than 1 standard deviation. Alternatively, "about" can mean a range of up to 20%, up to 10%, up to 5%, or up to 1% of a given value. Alternatively, particularly with respect to biological systems or methods, the term "about" or "approximately" can mean within a certain order of magnitude of a value, within 5 times, or more preferably within 2 times. If a specific value is described in the present application and claims, unless otherwise indicated, it should be assumed that the term "about" means within an acceptable error range for the specific value. The term "about" can have a meaning as commonly understood by one of ordinary skill in the art. The term "about" can mean ±10%. The term "approximately" can mean ±5%.

[0163] In the case of providing a range of values, it should be understood that unless the context clearly stipulates otherwise, each intermediate value between the upper and lower limits of the range is also specifically disclosed to the tenth of the lower limit unit. Each smaller range between any stated value or intermediate value in the stated range and any other stated value or intermediate value in the stated range are encompassed within the embodiments of the present disclosure. The upper and lower limits of these smaller ranges can be independently included or excluded within the scope, and each range including any limit, no limit or two limits in the smaller range is also encompassed within the disclosure, subject to any particularly excluded limit in the stated range. When the stated range includes one or two in the limits, the scope excluding any one or two of those included limits is also included in the disclosure.

[0164] Standard abbreviations may be used, eg, bp, base pair; kb, kilobase; pi, picoliter; s or sec, second; min, minute; h or hr, hour; aa, amino acid; nt, nucleotide; etc.

[0165] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Although any methods and materials similar or equivalent to those described herein can be used in the practice and testing of embodiments of the present disclosure, some potential and exemplary methods and materials are now described. Detailed description

[0166] We developed a method for inferring the methylation status of cfDNA molecules based on the use of cfDNA fragmentation patterns. The fragmentation pattern herein may involve (but is not limited to) fragment ends, cleavage patterns, fragment end motifs, genomic coordinates of fragment ends, and fragment sizes.

[0167] We have found a cutting pattern associated with CpG methylation status. This cutting pattern can be used to predict the methylation of a region at single-base resolution or on a per-fragment basis. In addition, it has been found that cfDNA terminal motifs (e.g., 3-mers or higher) can be used to predict the methylation of a region. Therefore, the cutting pattern of the fragment measuring the end in the window around the site can be used to infer the methylation of one or more sites in the region, the site in the genome, or the fragment. As another example, the fragment terminal motif can be used to infer the methylation level (e.g., regional methylation level) of the genomic region of interest. The fragment terminal motif can be defined by one or more nucleotides at the end of the free DNA fragment. As a further example, the cfDNA fragmentation feature can be used to infer the methylation state (i.e., the methylation pattern of the CpG sites present in the DNA molecule across the DNA molecule) at single-molecule resolution. For example, the terminal motif from a DNA molecule can be used to predict the methylation state of the entire fragment.

[0168] A further aspect of the present disclosure provides for determining the proportional concentration of clinically relevant DNA using a cleavage pattern. For example, methylation-related cfDNA fragment omics features can be used for, but are not limited to, tissue-derived analysis of plasma DNA. As described herein, various fragment omics features, such as cleavage patterns and terminal motifs, can be used. Certain sites can be selected for determining such proportional concentrations using fragmentation measurements. For example, high methylation and / or low methylation sites / regions of a specific tissue type can be used. As another example, sites of 5hmC enrichment or 5hmC deletion of a specific tissue type can be used. In addition, the technology can use fragmentation measurements (e.g., terminal motifs and cleavage patterns) to identify regions of 5hmC enrichment and 5hmC deletion of a specific tissue.

[0169] A further aspect of the present disclosure provides the detection of lesions. As an example, the whole genome or the genome containing CG motifs from tissue-specific (e.g., liver-specific) hypomethylated or hypermethylated sites / regions can provide diagnosis / classification of cancer (e.g., HCC) or other diseases that may involve specific tissues. The DNA of the object and / or pathogen (e.g., virus) can be used. For any type of DNA, sites with similar methylation levels can be used across different objects (with or without lesions), but wherein the objects have different cleavage patterns. As another example, sites / regions with 5hmC enrichment or 5hmC deletion of a specific tissue type can be used, for example, to identify lesions of a specific tissue type. Machine learning models can use such tissue-specific methylation patterns (e.g., hypermethylation and hypomethylation and 5hmC enrichment and 5hmC deletion) to distinguish different types of cancer. The free DNA fragments used for fragmentation measurement can be screened, for example, based on size. Therefore, lesions can be detected based on the amount of the terminal motifs of the free DNA fragments with a specific size.

[0170] Fragmentation information (fragment omics features) can be measured in various ways. As an example, high-throughput parallel sequencing or whole genome sequencing can be used. As another example, targeted sequencing methods can be used to determine the cfDNA fragment omics features from the region of interest for methylation pattern prediction. As an additional example, cfDNA fragment omics features can be detected using methods that are not based on sequencing, such as quantitative polymerase chain reaction (qPCR), real-time PCR, digital PCR (dPCR), droplet digital PCR (ddPCR), mass spectrometry, etc.

[0171] The exemplary use of a 5' terminal motif is provided as an example only. A 3' terminal motif may be used in addition to or in place of any of the embodiments described herein. I. Relationship between fragmentation and methylation

[0172] The present disclosure shows the relationship between the position of cutting DNA (e.g., as part of forming free DNA) and the methylation index (state) of the site in the genome. The position of cutting DNA can be defined as the cutting pattern. In another example, the position of DNA cutting can be represented by the terminal motif. A. Definition of cutting position

[0173] Figure 1The cut positions relative to CpG sites (also referred to as CG sites) according to embodiments of the present disclosure are shown. The horizontal line refers to a reference sequence, such as a portion of a reference genome, which contains two CpG sites. After sequencing, the resulting reads can be mapped to this region. The distance between the ends of the fragments and the positions relative to the CpG sites can be calculated. For example, the end of fragment 110 is exactly at the CG position and therefore has a distance of 0 (also referred to as position 0).

[0174] Fragment 120 ends one position to the left of CpG site 102. Since this fragment is cleaved one base before the CpG site, the distance is considered to be 1. In this way, the distance between the cleavage end and the CpG site can be calculated, and fragments with the same distance can be grouped together. As shown below, this distance is correlated with the methylation level.

[0175] Figure 2A The demonstration allows DNA fragments to be grouped based on the distance from the 5' end to the CpG site. If the first two nucleotides at the 5' end of the fragment are CG, the aforementioned distance is 0. If there is a nucleotide at the 5' end immediately before the CG, the aforementioned distance is 1, which corresponds to a fragment with an NCG motif, where N is any of the four bases. Sequenced CpG sites can be grouped into different categories based on their distance from the 5' end. Figure 2A The maximum distance shown is 8, but greater distances can be used. B. Cleavage position is associated with cleavage CpG methylation

[0176] To investigate whether cfDNA fragmentation patterns could be used to predict methylation, we investigated the relationship between methylation metrics and the CpG position relative to the 5' end of the fragment (i.e., the nucleotide distance between the CpG site and the 5' end). We analyzed bisulfite sequencing reads from eight plasma DNA samples from healthy subjects (median paired-end sequencing reads: 381 million (interquartile range: 353 million to 404 million)). The sequences were then aligned to a reference genome. The methylation density of each CpG site was determined using the methylation values ​​of the sequence reads at that CpG site.

[0177] The methylation index for each distance can be determined in various ways. For example, the number of methylated CpG sites covered by the DNA fragments in a group can be divided by the total number of sequenced CpG sites on the DNA fragments of each group. If there are 100 CpG sites on the fragment and 90 of them are methylated, the index will be 90. As another example, the methylation index can be determined as the number of fragments in the group that appear at high methylation sites (e.g., the methylation density at the site is greater than a specified percentage, such as 60%, 70%, 80%, etc.).

[0178] Figure 2B The methylation level is provided for each CpG site at a given downstream distance from the cleavage edge of the plasma DNA molecule (eg, the methylation level for a CpG site at a distance of 5 nt from the 5' end). Figure 2B The methylation index varies depending on the distance from the 5' end to the CpG site. There is a correlation between the methylation index of CpG sites and their distance from the 5' end. As can be seen, if the distance is 0, the methylation level in the CpG is highest. If the distance is 1, the methylation level is lowest. As the distance increases further, the methylation level becomes quite stable. When the CpG site is methylated, cleavage will occur more frequently at the methylated C, while if the CpG site is unmethylated, cleavage will be more random.

[0179] Thus, CpG sites with a distance of 0 had a higher probability of methylation compared to those with a distance of 1 [mean methylation index: 88.7% (range: 87.3% to 90.3%) vs. 69.1% (range: 58.2% to 76.9%)]. Because the methylation index differs depending on the distance between the 5' end and the CpG site, the fragmentation pattern can be used to predict the methylation level of those CpG sites.

[0180] The cleavage position is also related to the motif type. For example, if the fragment is cleaved exactly at the CG position, the terminal motif is CGN**** (where * represents a base and N is any of the four bases). If the fragment is cleaved at the base immediately preceding the CG, the terminal motif is NCG*** (where * represents a base), and so on. Therefore, there is a relationship between the motif type and the methylation level of the CpG site surrounded by the cleaved fragment.

[0181] To validate our observations, we further investigated the methylation levels in 256 possible sequences of different types of 4-mer motifs, namely four positions at the ends of the fragments.

[0182] Figure 3 A plurality of 4-mer terminal motifs according to embodiments of the present disclosure and the percentage of methylated CpG sites covered by fragments having the specific 4-mer terminal motifs are shown. Figure 3 In the figure, each dot represents a designated 4-mer motif. The individual motifs were grouped based on the distance of the cleavage site from the CG. As can be seen, the CGNN motif still shows the highest methylation level, while the NCGN motif shows the lowest methylation level, with NNCG falling in between.

[0183] As illustrated in 301, if the CpG site is methylated, it is highly likely to be cleaved right at the CG. If it is unmethylated, it is likely to be cleaved at the base before the CpG site in 302. This is why many fragments starting with CG are observed to be methylated, but many fragments starting with NCG are unmethylated.

[0184] Thus, some embodiments can use the amount of CGN motifs and / or NCG motifs to determine a methylation index for a region including a site or multiple sites. Other examples use cleavage patterns (described in more detail below) and the relationship between distance and methylation to determine methylation. CGN and NCG end motifs can be used to determine cleavage patterns, and cleavage patterns can include counts of CGN and NCG end motifs (also referred to as motifs). C. Probe-based counting of certain ends

[0185] In the examples above, sequencing was used to determine the amount of fragments ending at a specific distance from a CpG site or to determine the end motif of a fragment. Instead of sequencing, probe-based techniques can be used to measure the amount of CGNs (or longer motifs, such as CGNNs) relative to NCGs. Therefore, the frequency of CGNs and NCGs from a specific region can be detected by other methods, such as qPCR, digital PCR, and droplet digital PCR.

[0186] Figure 4 An example of a method for determining NCG and CGN motifs in a region of interest using droplet PCR according to an embodiment of the present disclosure is shown. As shown, cfDNA molecules 400 can undergo DNA end repair, A-tailing (adding A to the ends of the fragments), and common adaptor ligation as optional steps. DNA end repair can form blunt ends, so there is no overhang between the two strands. Common adaptors 401 can be added to the ends of the fragments.

[0187] The adaptor-linked molecules can be distributed, for example, into different reaction systems, such as droplets. A pair of PCR primers can be designed in the following manner: one primer (e.g., common forward primer 410) can bind to the common adaptor 401, and the other primer (e.g., region-specific reverse primer 420) can bind to a specific region of interest.

[0188] The DNA molecule will be amplified inside the reaction (e.g., droplet) by the pair of PCR primers. Two different fluorescent probes can be used, such as probe 430 for detecting CGN termination fragments and probe 440 for detecting NCG termination fragments. Fluorescent probes that are specific to specific terminal motifs (e.g., CGN or NCG terminal motifs) can be hydrolyzed and emit fluorescent signals, making it possible to detect the presence of specific motifs and to quantify specific motifs. For digital PCR, the number of reactions that are positive for specific terminal motifs can be counted and used to determine the amount of DNA fragments with the terminal motifs in the analyzed area. For real-time PCR, the intensity of each signal can be used as a measure of the amount of DNA fragments with specific motifs at the end. The two intensities can be compared with each other. D. Targeted sequencing of CGN and NCG

[0189] Another example technique for analyzing fragmentation is targeted sequencing, such as by amplifying DNA fragments having sequences corresponding at least in part to one or more regions of interest and / or using capture probes that select such DNA fragments. Certain DNA molecules having specific end motifs corresponding to specific distances from CpG sites to fragment ends can be amplified. Targeted sequencing methods can be used to determine cfDNA fragment omics features from regions of interest in a cost-effective manner. Targeted sequencing methods can increase the depth of regions of interest, thereby improving the accuracy of methylation analysis. DNA fragments containing CGN and NCG motifs can be enriched during sequencing library preparation. Amplification makes sequencing more efficient because most sequence reads will include CGN and / or NCG motifs.

[0190] Figure 5 An example workflow for preparing a sequencing library by selectively amplifying DNA containing NCG or CGN terminal motifs according to an embodiment of the present disclosure is shown. As shown, the cfDNA molecule 500 can undergo a process of DNA end pairing, A-tailing, and common adaptor ligation as an optional step, as described above. A pair of PCR primers can be designed in the following manner: one primer 510 can bind to the common adaptor region 501 (P1), and the other primer 520 or 530 can bind to the connection region between the DNA and the common adaptor (P2). By controlling the last base in the 3' end of P2 to be CG (primer 520) or NCG (primer 530), i.e., ACG, CCG, TCG, GCG, DNA fragments containing CG or NCG terminal motifs can be selectively amplified.

[0191] Those libraries will then be sequenced. Such sequencing reads will be used to analyze CGN and NCG end motifs, allowing determination of methylation levels, patterns (e.g., at different sites or regions), and differences. In addition, selective amplification of DNA fragments containing CGN or NCG end motifs can enrich for reads mapped to regions with different methylation levels, such as reads from different tissues within the body, from tumor versus non-tumor sources, or from fetuses versus pregnant mothers. II. Example Use Cases Using Cleavage Patterns and Terminal Motifs

[0192] cfDNA is a mixture of DNA molecules derived from different sources, which may have different methylation states in one or more CpG sites. Different methylation states will affect the plasma DNA fragmentation pattern during the release of DNA molecules into the plasma due to cell death (e.g., via apoptosis) or other mechanisms (e.g., active secretion). Different methylation states may also affect the clearance of free DNA molecules. Therefore, using cfDNA fragmentation patterns will allow the methylation state of CpG sites to be inferred. The effect on cfDNA fragmentation patterns caused by the methylation state on CpG will involve plasma DNA cutting across a series of genomic positions near or remote from the CpG site, which may be mediated by one or more DNA nucleases (e.g., DNASE1 and DNASE1L3).

[0193] Figure 6 A schematic diagram illustrating inference of methylation status based on fragmentation patterns according to embodiments of the present disclosure is shown. DNA from different tissues will have different methylation profiles. For example, blood cells will have different methylation profiles compared to liver at sites such as site 610. At these three CpG sites, blood cells will have different methylation levels, and liver tissue will have different methylation levels, i.e., UMU for blood and UMM for liver.

[0194] The two tissues will contribute to the free DNA in the plasma, making the free DNA methylation level different at each site. For example, if the first site is hypomethylated in both tissues, the methylation level in the free DNA will be very low. However, if both tissues are methylated, the methylation level of the free DNA will be very high. And if one is methylated and the other is not methylated, a partially methylated CpG site will be obtained. Figure 6 Such a scenario for methylation levels is shown in Methylation levels are shown as overall levels that can be measured in plasma according to the example methylation patterns depicted for blood cells and liver.

[0195] At the same time, both tissues (and other tissues in vivo) also promote their free fragment omics features into plasma. Therefore, if the first site is hypomethylated, there will be more unmethylated motifs (e.g., NCG) and the cutting will be more random; and at the second site, if both tissues are hypermethylated, there will be more methylated motifs (CGN). Therefore, the terminal motifs and cutting patterns will be a mixture of tissues, and this information can actually be used to infer the methylation status of each site based on its pattern analysis. This is shown in the section about cfDNA features, where site 605 has an unmethylated associated terminal motif 602, and site 608 has a methylated associated terminal motif 601.

[0196] Figure 6 Two methods are also shown to infer methylation status: method 620 uses cleavage patterns, and method 630 uses terminal motif patterns.

[0197] The cutting pattern can correspond to the amount of DNA fragments at the position of the end around the CpG site. In this example, the methylation pattern is used, and the site methylated in two tissues has the highest peak. For unmethylated sites, there is no peak. For highly methylated sites, there is a peak. For partially methylated sites (for example, methylated only in some tissues), there is a lower peak. As shown, the cutting pattern associated with the CpG site can be used to predict the methylation index of the plasma DNA molecules derived from different tissues (such as liver, lung, colon, small intestine, lymphocyte, neutrophil, etc.). The cutting pattern can be defined as, but not limited to, the amount of the fragment ends of each position (also referred to as cutting measurement window) across the multiple genomic positions analyzed around the CpG site. The number of genomic positions (called window size) can be, but not limited to, 1nt, 2nt, 3nt, 4nt, 5nt, 6nt, 7nt, 8nt, 9nt and 10nt or more. The amount of the fragment ends can be a standardized value, such as, but not limited to, the number of fragment ends divided by the sequencing depth at each position.

[0198] Method 630 can use terminal motif pattern to predict the methylation level in plasma.If the methylation level is low (for example, the first site shown), then there will be more terminal motifs related to unmethylation.If the methylation level is higher, then there will be more terminal motifs related to methylation.Using these two signals, it is possible to predict whether a CpG site or region is unmethylated, highly methylated or partially methylated.The terminal motif pattern of the genomic region being paid close attention to (that is, the regional terminal motif pattern) can be used to infer its corresponding methylation density (for example, the percentage of the sequencing CpG that is determined to be methylated).

[0199] Because methylation is tissue-specific and because fragmentomic features reflect methylation, methylation can further reflect tissue of origin. Fragmentomic features can be used directly (or inversely to determine methylation) to determine the proportional concentration of a specific tissue type (e.g., using tissue deconvolution if multiple tissue concentrations are measured), which can provide the proportional concentration of DNA from a specific tissue (e.g., clinically relevant DNA). Cleavage pattern and end motif analysis can be used to estimate the proportion of a specific tissue. Therefore, methylation-related cfDNA fragmentomic features (e.g., cleavage pattern and motif analysis) can be used to infer the proportion (proportional concentration) of cfDNA from different tissues of origin.

[0200] Some embodiments may also use fragmentomic signatures to infer the presence or absence of a pathology (eg, a disease such as cancer). III. Prediction of methylation sites based on fragmentation

[0201] This section describes a technique for predicting methylation indices in a genome using cleavage patterns at single CpG resolution. Cleavage patterns can be determined in a variety of ways, including using certain terminal motifs, such as CGN terminal motifs. Example definitions and results are provided below. Regardless of how the cleavage pattern is defined, such as the type of normalization used, the results are consistent. A. Normalization by total reads covering a position

[0202] The cleavage pattern is constructed based on the cleavage ratios across the genomic coordinates within the measurement window associated with the CpG site. The cleavage ratio at a position within the measurement window of interest can be calculated by the following formula: Thus, in this example, the cleavage ratio is defined as the number of ends at a position divided by the number of reads covering that site (sequencing depth).

[0203] Figure 7 The definition of a cleavage pattern using a cleavage ratio according to an embodiment of the present disclosure is shown. A cleavage measurement window 710 of width 11 is shown, but other widths can be used. A set of free DNA fragments 700 is shown above a reference sequence 720. The relative position is relative to the C of the CpG site. The row marked "end" shows the number of free DNA fragments whose ends are located at that specific base position. The row marked "depth" shows the number of free DNA fragments that overlap with that base position. The cleavage ratio is calculated using the equation shown.

[0204] The C of the CpG site is at position 0, corresponding to the CGN terminal motif. Positions -5 to -1 are upstream. Position -1 corresponds to the NCG terminal motif. A G is at position 1, and the other four are downstream, through position 5. These positions form the cleavage measurement window 710. Within this window, all free fragments are mapped. For each position, the number of termini at that position is calculated. This is displayed in the row labeled "Termini."

[0205] Then the depth of each position is calculated. For the A at position -5, there are two fragments mapped to this position, so the depth is two. Since the A at position -5 has one fragment ending at this position and a depth of two, the cutting ratio will be 0.5 (or 50%). 1. CG cutting mode

[0206] We analyzed the cleavage patterns of two groups of CG sites: putatively unmethylated and putatively methylated. Because the majority of cfDNA molecules in healthy participants originate from leukocytes (Sun et al. Proc Natl Acad Sci USA. 2015; 112: E5503-5512), we identified 1,000 putatively methylated CpG sites (β-value > 0.8) and 1,000 putatively unmethylated CpG sites (β-value < 0.1) based on the publicly available Illumina HumanMethylation450 BeadChip data from leukocytes (GSE40279). If a site was methylated in leukocytes, it was assumed to be putatively methylated. To investigate how methylation affects cfDNA fragmentation, 1,000 putative unmethylated CpG sites and 1,000 putative methylated CpG sites of cfDNA samples from eight healthy controls were analyzed for cleavage patterns, with a median number of paired-end sequencing reads of 381 million (interquartile range: 353 million to 404 million). Probe-based techniques, including targeted sequencing, can also be used as described herein. "Putative" means that methylation is derived from locus array information from publicly available databases.

[0207] Figure 8 A comparison of the cleavage patterns of methylated CpG sites 810 and unmethylated CpG sites 820 according to embodiments of the present disclosure is shown. The x-axis represents the nucleotide position relative to the CpG site within the measurement window. As an example, the window herein is defined as 5 nucleotides upstream and downstream of the C base in the CpG site (i.e., 5 nt) (i.e., a window size of 11 nt).

[0208] The y-axis represents the average cutting ratio. For each sample, the cutting pattern from the putative methylated CpG site or the putative unmethylated CpG site was merged by calculating the average cutting ratio of each position in the 11nt window. As shown, compared with the putative unmethylated CpG site (median cutting ratio: 0.54; range: 0.39 to 0.62), the cutting pattern of the putative methylated CpG site merged was shown in a higher cutting preference (median cutting ratio: 1.24; range: 0.91 to 1.37) at position '0' (i.e., the cytosine nucleotide of the CpG site in question) (P value=0.00016, Mann-Whitney U test (Mann-Whitney U test)). Therefore, once C is methylated, the cutting is more likely to be at C; and when C is unmethylated, the cutting is less likely to be at C. 2. Cutting modes for other sequence backgrounds

[0209] In addition to methylation status, sequence context may also affect cleavage patterns. To investigate the effects of sequence context, pooled cleavage patterns associated with methylated and unmethylated CpG sites were determined from a set of cleavage measurement windows where position '-1' corresponds to a cytosine nucleotide, i.e., a "CCG" subsequence (6,928,652 measurement windows in total).

[0210] Figures 9 and 10 Shows the cleavage patterns of windows associated with different sequence contexts (including CCG and methylation status) according to embodiments of the present disclosure. For example, this data was generated using bisulfite sequencing results from a healthy control sample (391 million paired-end sequencing reads).

[0211] Figure 9 Shown are cleavage patterns for windows containing C, C, and G nucleotides at positions -1, 0, and 1. Black line 910 represents the cleavage pattern for windows containing a methylated C at position 0, and gray line 920 represents the cleavage pattern for those windows containing an unmethylated C at position 0. In this example, the methylated state (represented by M) is defined by those CpG sites with a methylation index greater than 70%, while the unmethylated state is defined by those CpG sites with a methylation index less than 30% (represented by U).

[0212] The cleavage rate at position '0' at methylated CpG sites (0.90%) appears to be much higher than those at unmethylated CpG sites (0.51%). When C is methylated, the likelihood of cleavage at C is greater; when C is unmethylated, the likelihood of cleavage at that C is decreased.

[0213] Figure 10The cleavage patterns of 615,465 windows comprising C, G, C and G nucleotides at positions 0, 1, 2 and 3 (i.e., the CGCG subsequence) are shown. According to the methylation pattern of the two CpG sites of the CGCG subsequence, the cleavage patterns are divided into four groups, i.e., "MM", "MU", "UM" and "UU". The black solid line 1010 represents the cleavage pattern of the window comprising methylated C at position 0 and position 2, and the gray dotted line 1020 represents the cleavage pattern of the window comprising unmethylated C at position 0 and position 2. The black dotted line 1030 represents the cleavage pattern of the window comprising methylated C at position 0 and unmethylated C at position 2, and the gray solid line 1040 represents the cleavage pattern of the window comprising unmethylated C at position 0 and methylated C at position 2.

[0214] Compared with the "UU" group (cutting ratios were 0.53 and 0.42), for the "MM" group, relatively high cutting ratios were observed in the two cytosine positions of the CGCG subsequence (cutting ratios were 0.86 and 0.89). When the methylation status of the 2 CpG sites was different (i.e., the "MU" group and the "UM" group), relatively high cutting tended to occur at the methylated cytosine. These data show that the cutting pattern is related to the methylation pattern across a series of different CpG sites. Therefore, the cutting pattern can be used to infer the methylation pattern across a series of different CpG sites. 3. CGN / NCG ratios in different sequence backgrounds

[0215] For example, in Figures 2 to Figure 3 In , it has been shown that CGN and NCG motifs can be used to reflect the methylation status of CpGs immediately adjacent to the cleavage site of interest. Figure 10 The cleavage pattern for two CpGs that are close to each other (ie, in tandem) is shown.

[0216] In order to understand the relationship between the cfDNA cutting pattern and the methylation status of several adjacent CpGs, the CGN / NCG motif ratios of different combinations of methylation states of those molecules with multiple adjacent CpG sites were analyzed. The CGN / NCG motif ratio corresponds to the ratio of the amount of DNA fragments with the CGN motif to the amount of DNA fragments with the NCG motif. Here, we define adjacent CpGs as those CpG sites that are within a size range of 75bp but not in series. In other embodiments, adjacent CpGs can be defined as CpG sites within a range of (but not limited to) 5bp, 10bp, 20bp, 30bp, 40bp, 50bp, 100bp, 200bp, 500bp, 600bp, 1000bp, etc. In another embodiment, tandem CpGs can be used. Because the majority of cfDNA molecules (approximately 93.5%) contain no more than three CpG sites within 75 bp per molecule, the CGN / NCG motif ratios of those molecules with two and three CpG sites, respectively, were analyzed across different combinations of methylation status.

[0217] Figure 11 Schematic diagram showing cfDNA grouping at adjacent CpG sites. Cleavage site 1105 is to the left of the 5' end. In this example, range 1110 is 75 nt. The first two examples show DNA fragments with two CpG sites. The second two examples show DNA fragments with three CpG sites. The right side shows different combinations of methylation status.

[0218] For cfDNA molecules with two CpGs, there are a total of four combinations of methylation status. One combination can be the case where a methylated CpG at the 5' end is followed by a methylated CpG (represented by "MM," where "M" represents a methylated CpG and '-' represents any one or more nucleotides). Other combinations can be "MU," "UM," or "UU," where "U" represents an unmethylated CpG. For cfDNA molecules with three CpGs, there are a total of eight combinations of methylation status: "MMM," ​​"MMU," "MUM," "MUU," "UMM," "UMU," "UUM," and "UUU."

[0219] FIG. 12A to FIG. 12B The effect of methylation patterns with multiple adjacent CpGs on the CGN / NCG motif ratio is shown. FIG. 12A to FIG. 12BAs shown in Figure 3, the CGN / NCG motif ratio is significantly higher in those cfDNA molecules that start with a methylated CpG at the 5' end (i.e., "MM," "MU," "MMM," ​​"MMU," "MUM," and "MUU") compared to molecules that start with an unmethylated CpG at the 5' end (i.e., "UM," "UU," "UMM," "UMU," "UUM," and "UUU"). This is similar to the behavior of single CpG sites, suggesting that the main effect is the terminal motif at the cleavage site of interest.

[0220] Interestingly, the cleavage of molecules starting from an unmethylated CpG at the 5' end appears to be relatively enhanced by the presence of methylation of the adjacent CpG. For example, for those molecules with two CpG sites, the "UM" group showed an approximately 27.9% increase in the CGN / NCG motif ratio compared to the "UU" group (P value = 0.018) ( Figure 12A ).

[0221] In addition, for those cfDNA molecules with three CpGs starting from an unmethylated CpG at the 5' end, the CGN / NCG motif ratio tended to gradually decrease as more adjacent CpGs became unmethylated. Compared with "UMU", "UUM" and "UUU", the "UMM" group showed an increase in the CGN / NCG motif ratio of approximately 20.6%, approximately 25.0% and approximately 52.2%, respectively ( Figure 12B ).

[0222] Therefore, the methylation status of CpGs immediately adjacent to the cleavage site of interest showed a more significant effect on cfDNA cleavage than those CpG sites further away from the cleavage site. In addition, the data also suggest that CpG cleavage at the 5' end may be at least partially affected by the methylation status of adjacent CpGs. This finding may have several potential applications.

[0223] In one embodiment, 5' end CGN and NCG motifs can be used to infer the methylation index of several adjacent CpGs in the reference genome. Although it is best to use the end motif at the cleavage site being paid attention to, the number of DNA fragments at the end may be insufficient, for example, because the CpG site may be located in the nucleosome, so the chance of DNA enzyme cutting at this type of site is less. In such cases, the methylation state can be inferred based on the CGN / NCG at the upstream CpG site. For example, as shown, a motif ratio of about 2 will indicate that the next CpG site is methylated, while a motif ratio of about 1.7 will indicate that the next CpG site is not methylated. Therefore, the motif ratio at the cleavage site can determine the methylation index at the cleavage site, and can also determine the methylation index at one or two CpG sites downstream.

[0224] In another embodiment, cfDNA cleavage patterns associated with one or more adjacent CpGs can be used to enhance the ability to diagnose pathological lesions. For example, cfDNA cleavage patterns associated with many CpGs that show the same methylation pattern within a certain nucleotide distance can be used to assist in the selection of diagnostic markers. As another example, fragment omics features derived from cancer-specific hypomethylated markers can be used to diagnose cancer. In some cases, using a cleavage pattern that spans CpG sites with adjacent hypomethylated CpGs can outperform using CpGs with adjacent hypermethylated CpGs. B. Normalize by total end points in the region

[0225] Another example of normalization uses the number of fragments whose ends are located in the region surrounding the site. This region can be the same as or different from the window used to determine the cleavage pattern.

[0226] Figure 13 Figure 1 illustrates normalization of cleavage patterns using cleavage density according to embodiments of the present disclosure. Black line 1305 shows a reference genome containing CG 1301. As with the example using cleavage ratios, to infer the methylation status of a CpG site, one can calculate how many fragments have ends located at that CpG site. For normalization, the number of ends at each position is divided by the total number of ends in the entire region. This is another example technique that can be used to normalize the amount of DNA fragments with a specific end motif or ending at a specific position.

[0227] Similar analyses are performed to determine the patterns of various CpG sites with respect to cutting density, as was done for cutting ratios. For example, one CG, two CGs, and three CGs are analyzed. Sequencing analysis uses merged data from multiple samples to obtain a higher sequencing depth. If the methylation index is greater than a first critical value (80% in this example), the site is methylated, and if the methylation index is less than a second critical value (30% in this example), the site is not methylated. If the methylation index of a site is between the critical values, it is not used for this example. In other embodiments, an intermediate classification can be used.

[0228] Figure 14 Cleavage patterns are provided using cleavage densities for various methylation combinations of two CpG sites according to embodiments of the present disclosure. The CpG sites are tandem in this example, as Figure 10 There are four groups based on the methylation status of the first CG and the second CG. Here, two dotted lines 1403 and 1404 represent the positions of these two cytosines.

[0229] It can be seen that in the upper two pattern graphs, the first C 1401 is not methylated, and therefore there is no peak at C 1201. However, in the lower two pattern graphs, the first C is methylated, so there is a significant increase in the cutting density from -1 to 0 position. For the second and fourth pattern graphs, the second CG is methylated. In these patterns, there is a peak at the second C 1402. For the first and third pattern graphs, the second CG is not methylated, so the peak is lower or disappears. Regarding the cutting density 1405, it has a high value, but there is no relative increase. The relative increase is more important for inferring the methylation status, rather than the exact numerical value at the position. These results are consistent with Figure 10 The results using the cutting ratio are consistent with those in

[15] .

[0230] Figure 15 Cleavage patterns using various methylation combinations of three CpG sites according to embodiments of the present disclosure are provided. Different patterns correspond to different combinations of methylation states at different sites. As can be seen, the top two patterns have the first CG methylated while the bottom two patterns are unmethylated. We also observe significant peaks 1503 and 1505 at the first C1501 in the top two patterns. However, the peak for the first C1501 disappears in the bottom two patterns.

[0231] In the first and third patterns, the second CG is methylated. We observe a peak 1504 for the second C1502. In the second and fourth patterns, the second CG is unmethylated, and the peak is reduced or disappears.

[0232] In the first three patterns, the third CG is methylated and a peak is observed. In the fourth pattern, the third CG is not methylated and therefore the peak is very low.

[0233] Figure 16 The cutting density of the window including CCG and methylation status according to the embodiment of the present disclosure is shown. The region containing CCG is analyzed because DNASE1L3 prefers to cut at CC or at the end of CCC. But if CG is methylated, we assume that it will prefer to cut exactly at this CG (between CC). Therefore, if it is methylated, peak 1601 is observed. Therefore, if CG is methylated, it prefers to cut between two Cs. If CG is not methylated, the peak disappears. These results are consistent with Figure 9 The results using the cutting ratio are consistent with those in

[15] .

[0234] Figure 17 Shown is a comparison of cleavage patterns between methylated and unmethylated CpG sites using cleavage density according to embodiments of the present disclosure. Figure 17 and Figure 8are comparable and show similar results, but use a different definition of cutting mode, i.e. using cutting density as opposed to cutting ratio. Figure 8 Similarly, we selected putative hypermethylated CpG sites or hypomethylated CpG sites from the public leukocyte methylation array data. We randomly selected 1,000 CpG groups that were methylated in leukocytes (β value > 0.8) and 1,000 CpG sites that were unmethylated in leukocytes (β value < 0.1).

[0235] Each red line 1720 represents a sample with one thousand methylated CpG regions, and each blue line 1730 represents a sample with one thousand unmethylated CpG sites. We observe a significant peak in the red line, which means that in highly methylated regions, the methylation status can be inferred from this peak. And for the blue line, the peak here is reduced or even disappears. Therefore, if the CpG is methylated, there will be a peak for the fragment whose end is located at position 1701. This peak will be seen in the schematic diagram 1707 representing the cleavage pattern. When C is unmethylated, the preference for this CG end will disappear, as shown in the schematic diagram 1709. The results show that the methylation status can be predicted at a single CpG resolution by the cleavage pattern. C. Results of using support vector machines

[0236] In some embodiments, machine learning (ML) technology can be used. Such technology can use two or more quantities in the cutting measurement window, and can include a complete cutting pattern in the window around the CpG site. Machine learning technology can also use sequence background, such as the example of the complete sequence or k-aggregate, which can be recorded in various ways. Sequence background can also be used outside machine learning, such as other relative increases from -1 to 0 position can be different according to the base before C. Machine learning can identify such patterns in any position in the feature vector input for the ML model. Various ML models can be used. This section uses support vector machine (SVM) to classify sites as hypermethylation or hypomethylation. 1. Use cutting mode only

[0237] In some embodiments, the cleavage pattern is constructed using the cleavage ratio from an 11nt cleavage measurement window centered on the CpG site. The cleavage pattern from the Watson and Crick strands is used to train a machine learning model that is used to infer the methylation state of the CpG site at the center of the window. In one embodiment, a support vector machine (SVM) is used to predict whether the methylation index at the CpG site is greater than 95% or less than 20% based on the cleavage pattern associated with the CpG site.

[0238] Figure 18A workflow for classifying CpG sites as hypermethylated or hypomethylated using SVM according to an embodiment of the present disclosure is shown. For the training set, 33,147 hypomethylated CpG sites and 33,147 hypermethylated CpG sites in the reference genome were used. The actual classification of each site was determined using bisulfite sequencing. In this example, hypomethylation is a methylation index below 20%, while hypermethylation is a methylation index above 95%. Half of the sites are used for training and the other half for testing. The trained model can predict whether a CpG site is hypomethylated or hypermethylated based on the cleavage pattern. The input feature vector is the cleavage ratio at 11 positions, as shown at each site. Feature vector 1810 is an example of such a feature vector.

[0239] Figure 19 The performance of single CpG site methylation state prediction using a support vector machine (SVM) based on cleavage patterns according to embodiments of the present disclosure is demonstrated. Two graphs show data from the Watson strand and the Crick strand. When using only data from the Watson strand or the Crick strand, an AUC of approximately 0.8 can be achieved. In this example, the ratios were normalized to the depth of cleavage in the cleavage pattern using sequencing at each position.

[0240] In some embodiments, the methylation pattern of the CpG site in the Watson chain is symmetrical with the methylation pattern of the corresponding site in the complementary Crick chain. The cutting pattern from the Watson chain and the Crick chain can be merged to train a machine learning model for inferring the methylation state of the CpG site in the window center. Therefore, in some embodiments, the cutting pattern from each chain is used. Therefore, the feature vector can have 22 values ​​in this example of the 11 positions in the measurement window. In most cases, the two chains share the same methylation state. Therefore, the methylation state of two CpG sites can be inferred using the chain.

[0241] Figure 20 The performance of using a support vector machine (SVM) to predict the methylation status of a single CpG site based on the cleavage pattern from two chains according to an embodiment of the present disclosure is demonstrated. An AUC of approximately 0.89 was achieved in the test data set, which appears to be superior to any model based on data from a separate Watson chain or Crick chain (P value < 0.0001, DeLong test). If the cutoff value of the output probability based on SVM is set to 0.5 for determining whether a CpG site is methylated, a sensitivity of 83.8% and a specificity of 77.2% can be achieved. If the cutoff value of the output probability based on SVM is set to 0.55 for determining whether a CpG site is methylated, a sensitivity of 81.4% and a specificity of 80.2% can be achieved. These data show that it is feasible to use a machine learning algorithm to predict methylation status using a cleavage pattern.

[0242] The sequencing depth (e.g., 10×) used for training samples can be higher than the sequencing depth used for testing samples. 2. Use Sequence Background

[0243] The sequencing context surrounding CpG sites can be analyzed to improve model performance. For example, a training dataset containing cleavage patterns associated with methylated and unmethylated CpG sites can be divided into four categories based on the type of nucleotide at the '-1' position of the Watson chain. The sequence context can be exploited by training different models: one for each of the four sequence contexts.

[0244] Figures 21A to 21B The performance of single CpG site methylation status prediction using support vector machine (SVM) based on cleavage patterns in different sequence contexts according to embodiments of the present disclosure is demonstrated. Using sequence context can improve accuracy.

[0245] Figure 21A Combined data from the Watson (W) and Crick (C) chains are shown. The different lines represent ROC curves when different subsequences are used at the -1, 0, and 1 positions within the measurement window of the Watson chain. C at the -1 position provides the greatest accuracy. For this embodiment, there are essentially four different models: models for A, C, G, and T on the Watson chain.

[0246] like Figure 21A As shown in , when focusing on the cutting measurement window with C, C and G nucleotides (i.e., CCG subsequence) at positions -1, 0 and 1 in the Watson chain, the AUC increased from 0.89 to 0.94 (P value < 0.0001, DeLong test), thus demonstrating the improvement of model performance. The use of cutting measurement windows with ACG, GCG and TCG subsequences at positions -1, 0 and 1 in the Watson chain will result in AUC values ​​of 0.86, 0.91 and 0.89, respectively. Therefore, for the two sequence backgrounds (CCG and GCG), the AUC value is 0.86, 0.91 and 0.89, respectively, relative to the AUC value. Figure 20 The results for were improved, the same results were found for TCG, but decreased for ACG.

[0247] Figure 21B Combined data from the Watson (W) and Crick (C) chains with CCG subsequences at positions -1, 0, and 1 within the measurement window from both chains are shown. Thus, using sequence context from both chains improves the AUC from 0.94 to 0.97. If both chains and all four bases were extended, there would be 16 different models.

[0248] Thus, in some embodiments, the use of cleavage patterns from windows with CCG subsequences at positions -1, 0, and 1 in both the Watson and Crick chains further enhanced model performance, reaching an AUC of 0.97. These results suggest that selective analysis of cleavage patterns based on sequence context and a combination of Watson and Crick chain data synergistically improves model performance in distinguishing methylated from unmethylated CpG sites. 3. Windows with different widths

[0249] exist Figure 20 In the example of , a measurement window of 11 bp is used to determine whether a given site in the genome is hypermethylated or hypomethylated. For any of the embodiments described herein, windows of various widths can be used, such as 2, 3, 4, 6, 7, 8, 9, 10, 11, and 20. The results of these widths are shown in Table 1. Table 1 summarizes the AUC for methylation prediction by SVM models with different window sizes. AUC (overall) by Figure 20 The method used in the analysis, and the other four Figure 21A The method used in uses the sequence context analysis of the reference genome at position -1. The measurement uses fragments from the Watson chain (W) chain and the Crick (C) chain, but the measurement can only use one chain. The sequence context is defined for the Watson chain, but can be defined according to the Crick chain and according to both. Additional or alternatively, other sequence contexts (except the -1 position) can be used, for example, at positions -3, -2, 1, 2 and 3. Table 1: AUC of methylation prediction by SVM models with different window sizes.

[0250] For a 1 bp window at position 0, at a CpG site, the number of DNA fragments whose ends are located at the C of the CpG site is used to determine the cleavage ratio. For this example, the sequencing depth is used to determine the cleavage ratio, as described in Section III.A. The SVM model can receive the cleavage ratio of the Watson chain and / or the cleavage ratio of the Crick chain as input. Other machine learning models can be used, such as decision trees or neural networks.

[0251] In some embodiments, the present invention provides the method for the cutting of DNA fragments of described embodiment 1,2 or 3.When only determining a cutting ratio (for example, only a chain), the first cutoff value / critical value can be used for distinguishing hypermethylation from non-hypermethylation, and the second cutoff value / critical value can be used for distinguishing hypomethylation from non-hypomethylation.When using extra cutting ratio (for example, as a part for cutting pattern), extra cutoff value / critical value can be used.Cutting ratio can take various forms as described herein, for example, by the number of the DNA fragmentation of the number of covering position, the number of the DNA fragmentation in the region around this site or the number of the DNA fragmentation at one or more specific positions around this site by end and carry out standardization.Any one in this type of standardization can be used for cutting pattern.

[0252] For a 1 bp window at position -1, the cleavage ratio is determined at position -1 to the left of the CpG site. This position corresponds to the NCG end motif. For an AUC column using ACG, the DNA fragment will have ACG at the end, and so on for other AUC values ​​using the respective sequence context.

[0253] For the 1 bp window at position 1, the cleavage ratio is determined at position 1 to the right of the CpG site.For each of the remaining measurement windows, the specific position used is identified in the column for the window size.

[0254] The data in Table 1 were generated using the 33,147 hypomethylated CpG sites and 33,147 hypermethylated CpG sites described above. In this example, hypomethylation is defined as a methylation index below 20%, and hypermethylation is defined as a methylation index above 95%. D. Results of using convolutional neural networks in a sequence context

[0255] In some embodiments, a machine learning model (e.g., a neural network) can consider the sequence context and cleavage pattern (e.g., more than one cleavage density or cleavage ratio) surrounding the CG site to predict methylation. In one embodiment, a convolutional neural network (CNN) is implemented to determine whether a CpG site in the genome is methylated in order to synergistically utilize the sequence context and cleavage pattern present in the measurement window.

[0256] Figure 22 A schematic diagram showing the use of a CNN model for methylation state prediction based on cleavage pattern and sequence context according to an embodiment of the present disclosure. A methylated state can be defined by a methylation index exceeding 95% (or other values, such as those described herein), while an unmethylated state can be defined by a methylation index below 20% (or other values, such as those described herein). For illustrative purposes, for Watson chains, the upstream 5nt (e.g., ATCTG) and downstream 5nt (e.g., GAGTA) of the cytosine at the analyzed CpG site are presented as 5'-[ATCTG]C [GAGTA]-3'. The relative positions of this sequence correspond to -5, -4, -3, -2, -1, 0, +1, +2, +3, +4, and +5, respectively. The center position "0" corresponds to the cytosine at the CpG site analyzed for methylation status.

[0257] The fragment cutting ratio of each individual position can be constructed into 2-dimensional (2D) matrix 2220 according to sequence background. For example, for the position-1 corresponding to base guanine (" G "), the cutting ratio (1.40) relevant to " G " is filled in the corresponding lattice between "-1 " column and " G " row. The remaining rows corresponding to " A ", " C " and " T " in Watson chain are filled with " 0 ". The cutting pattern and sequence background (' 5-[TTACT] C[GCAGA]-3 ') derived from Crick chain are also processed similarly. The data matrix from Watson chain and Crick chain can be combined for downstream analysis. Therefore, this type of data matrix can contain both sequence background and cutting pattern information. Use those multiple data matrices relevant to known methylation status, i.e., methylated CpG sites (output value is " 1 ") and unmethylated CpG sites (output value is " 0 ") to train CNN model.

[0258] As shown, the cutting ratio of two chains is calculated. Matrix includes base (nucleotide), position and cutting ratio (using unique heat encoding in this example). For example, at -5 in the W chain, the first base is A, and the cutting ratio is about 0.25. Two matrices can be combined into a combined matrix 2230. This combined matrix can be the input metric of the CNN model, which has a convolutional layout to extract the features of these matrices. The extracted features can be combined into linear data to calculate the possibility of CpG site methylation or unmethylation. An AUC of about 0.96 can be achieved.

[0259] In one embodiment, the CNN model uses two one-dimensional (1D) convolutional layers, each with 64 filters of kernel size 4, but other hyperparameters can be used, such as 50 to 100 filters or kernel sizes of 3 to 6. The activation function of the rectified linear unit (ReLU) is used for those convolutional layers. A batch normalization layer is then applied, followed by a dropout layer with a dropout rate of 0.5. A flattening layer is further added, followed by a fully connected layer comprising 128 neurons using the ReLU activation function. Finally, an output layer with one neuron is applied, and a sigmoid activation function is used to obtain a probability score (i.e., a methylation score) for CpG site methylation. The program of the CNN model is implemented based on the PyTorch machine learning framework (https: / / pytorch.org / ). In various embodiments, the number of convolutional layers can be 1, 3, 4, 5, 6, 7, 8, 9, from 10 to 20, from 30 to 40, from 40 to 50 or more. The number of filters in each convolutional layer can be from 1 to 10, from 10 to 20, 20 to 30, 30 to 40, 40 to 50, 50 to 60, 60 to 70, 70 to 80, 80 to 90, 90 to 100, 100 to 150, 150 to 200 or more. The kernel size of the filter can be 2, 3, 4, 5, 6, 7, 8, 9, 10, from 10 to 20 or more. The CNN can also include multiple hidden layers containing multiple nodes. The first layer of the multiple hidden layers is coupled to the input layer. The CNN can further include an output layer, which is coupled to the last layer of the multiple hidden layers and is configured to output an output data structure.

[0260] The optimal parameters for the CNN model were obtained when the total prediction error between the output score obtained by the sigmoid function and the desired targeted output (binary value: 0 or 1) was minimized by iteratively adjusting the model parameters. The total prediction error was measured using the cross-entropy loss function in a deep learning algorithm (pytorch.org). The model parameters learned from the training dataset were used to analyze the test dataset to output a probability score (referred to as a methylation score in this study) that indicates the likelihood that a CpG site is methylated.

[0261] The methylation score can be within the range of 0 to 1. In one embodiment, a methylation score of 1 indicates that a CpG site can be considered as methylated. A methylation score of 0 can indicate that a CpG site can be considered as unmethylated. In some embodiments, if a specific CpG site is classified as "methylated" or "unmethylated" based on the cutoff value of the methylation score, the cutoff value can be used for classification. The possible values ​​of the cutoff value include 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90% or 95%. The methylation score of a CpG site is greater than a predefined cutoff value and can be classified as "methylated", while the methylation probability of a CpG site does not exceed a predefined cutoff value and can be classified as "unmethylated". The required cutoff value can be obtained from, for example, a receiver operating characteristic (ROC) curve analysis.

[0262] Figure 23A The performance of single CpG site methylation state prediction using a CNN based on cleavage pattern and sequence context according to embodiments of the present disclosure is shown. The solid line represents the performance of the model trained using both cleavage pattern and sequence context, and the dashed line represents the performance of the model trained using sequence context alone. Sequence context alone was trained only on the sequence pattern of the reference genome without any fragmentation information.

[0263] Based on the methylation score, the CNN model can achieve an AUC of 0.96, which is better than the SVM-based model (0.89) that does not include sequence background (P value < 0.0001, DeLong test). And this AUC is used for all sequence backgrounds. If only sequence background is used in the model but no cutting pattern is used, the performance will drop to an AUC of 0.79. These results indicate that the combined use of cutting pattern and sequence background will significantly improve the model performance of determining methylation status. In other embodiments, cutting pattern and sequence background can be used to infer methylation indicators (e.g., specific percentages or percentage ranges) rather than binary values ​​of high or low methylation.

[0264] In some embodiments, the machine learning model may include (but is not limited to) convolutional neural networks (CNNs), linear regression, logistic regression, deep recurrent neural networks (e.g., fully connected recurrent neural networks (RNNs), gated recurrent units (GRUs), long short-term memories (LSTMs)), transformation-based methods (e.g., XLNet, BERT, XLM, RoBERTa), Bayesian classifiers, hidden Markov models (HMMs), linear discriminant analysis (LDA), k-means clustering, density-based clustering with noise (DBSCAN), random forest algorithms, adaptive boosting (AdaBoost), extreme gradient boosting (XGBoost), support vector machines (SVMs), or a composite model including one or more of the above models. E. Results using various locations

[0265] As shown above, the cleavage distance reflects the methylation index, e.g. Figures 2A to 2B and Figure 3 , specifically distances 0 (corresponding to the terminal motif being CGN) and 1, where distance 1 corresponds to position -1 (corresponding to the terminal motif being NCG) and +1 (corresponding to the cleavage site between C and G). The cleavage ratios at various positions (e.g. determined using the amount at a particular position) also show differences between methylation levels, e.g. Figures 8 to 10 and Figure 17 Additional data are provided in this section showing that the amount of free DNA molecules ending at the first position of a CpG site can be used to distinguish differential methylation levels, where the first position is one of the -1 to +1 positions. As shown, position -1, position 0, or position +1 can be used.

[0266] Figure 23B A comparison of the sum of the average cleavage ratios at positions -1 and +1 between hypermethylated and hypomethylated CpGs in 39 ucfDNA samples is shown. The sum of the average cleavage ratios of 18,280,640 hypermethylated CpG sites and 2,576,102 hypomethylated CpG sites of 39 healthy control urine cfDNA samples was analyzed, with a median number of paired end sequence reads of 75.8 million (interquartile range: 60.5 million to 155 million). Each point represents an object. As shown, the cleavage ratios at positions -1 and +1 are different between hypermethylated sites and hypomethylated sites. Such data show that methylation can be predicted without using fragment information at position 0.

[0267] Figure 24AThe performance of the CNN model for single CpG site methylation state prediction based on the proximal cleavage pattern and sequence context excluding position 0 is shown. The sequence context is used as described above. In the absence of position 0, the combined matrix of the + / - 5 base window will have 10 columns instead of 11 columns, as shown in Figure 2. Figure 22 The AUC of 0.91 was achieved, which is lower than the 0.96 AUC using position 0, but still much higher than using the sequence background alone (AUC = 0.79). Figure 23B Same dataset.

[0268] As mentioned above, any of the positions -1 to +1 can be used individually. The amount of fragments ending at a particular position relative to the CpG site can be used, although some embodiments may use a ratio (e.g., a CGN / NCG ratio), or other embodiments may use a vector of cleavage ratios at multiple positions.

[0269] Figure 24B Demonstrate the performance of single CpG site methylation status prediction using CNN based on cleavage patterns and sequence context excluding positions -1, 0, and 1. Figure 24B It shows that a set of continuous or discontinuous locations within the cut measurement window can be used to form the data structure for input into the machine learning model for training. However, if some locations contribute significantly to the classification, for example, -1, 0, and 1, performance may be reduced.

[0270] Figures 25A to 25F Box plots showing the differences in average cleavage ratios at various positions in plasma and urine (ucfDNA) between hypermethylated and hypomethylated CpGs. Figure 25A Shown are the differences in the cleavage ratio at position -1 between hypermethylated and hypomethylated CpGs in plasma. Figure 25B The differences in cleavage ratios at position 0 in plasma are shown. Figure 25C The difference in cleavage ratio at position +1 in plasma is shown. The lower box plots indicate the difference in cleavage ratio at position -1 in ucfDNA ( Figure 25D )、Position 0( Figure 25E ) and position 1( Figure 25F ). Differences in the mean cleavage ratios of the two samples were significant if p < 0.001. Position 0 provided the best discrimination. Position -1 discriminated better for plasma than for urine, and position +1 discriminated better for urine than for plasma. F. Determining quantitative methylation density

[0271] In the above, examples are given of classification as hypermethylation (e.g., methylation density greater than X% (e.g., 80%)) or hypomethylation (e.g., less than Y% (e.g., less than 30%)). Some embodiments can use a calibration function (e.g., as described in the terminology section) to determine a narrower range or estimate a specific value. For example, the methylation level of a site can be determined using a methylation correlation analysis. Cleavage information, such as any of the cleavage parameters described herein, such as cleavage density or cleavage ratio, can also be determined.

[0272] The measured site methylation level and corresponding cutting information / parameter (such as cutting ratio or cutting density) can form calibration data points, wherein the cutting parameter is the calibration value corresponding to known / measured methylation level. Taking into account the relationship between cutting and methylation, when the cutting parameter changes, the methylation level will also change. When the cutting parameter changes, the calibration function can limit the proportional change of methylation level. In this way, when measuring the cutting parameter of a new sample (such as, sequencing under not carrying out methylation correlation analysis), the cutting parameter can be compared with the calibration function (such as with the calibration value corresponding to known methylation level). In one embodiment, cutting information / parameter (such as, cutting pattern or terminal motif) can be combined with methylation correlation analysis (such as, using an overall kinetic model for the data generated by single molecule real-time sequencing, such as in U.S. Publication No. 2021 / 0047679) to improve the final classification accuracy of measuring DNA methylation analysis.

[0273] Calibration data points can be used to train a model, such as a regression model. The regression model can be used to determine a quantitative value, which can be a range, such as a range of 5% (e.g., a density between 40% and 45%), 10%, 15%, 20%, 25%, or 30%. Regression can provide a calibration function. In some embodiments, the likelihood of an ML model (e.g., a neural network) can be used as a methylation indicator.

[0274] Figures 26A to 26C Shown is a comparison of various techniques for determining quantitative values ​​of methylation levels (eg, density) of CpGs with different methylation levels in eight plasma DNA samples. Figure 26A The average CGN / NCG motif ratio for a set of CpG sites with various ranges of methylation density is shown. Specifically, ranges of 20% width are shown as follows: 0 to 20%, 20 to 40%, 40 to 60%, 60 to 80%, and 80 to 100%. Figure 26B The average cleavage ratio at position -1 relative to the CpG site is shown. Figure 26C The average cleavage ratio at position 0 of the CpG site is shown.

[0275] In the whole genome methylation sequencing (WGBS) data of plasma DNA samples, CpGs were divided into five groups according to a certain range of methylation index, namely 0 to 20%, 20 to 40%, 40 to 60%, 60 to 80%, and 80 to 100%. In the same sample, the average CGN / NCG motif ratio of CpG in each group was calculated ( Figure 26A ), the average cutting ratio at position -1 ( Figure 26B ) and the average cutting ratio at position 0 ( Figure 26C ) and plot it. Each point represents an object.

[0276] All three parameters consistently showed a relationship with methylation density (increase or decrease). The CGN / NCG motif ratio gradually increased as the methylation index increased. The cleavage ratio at positions -1 and 0 was negatively correlated and positively correlated with different groups of methylation index ranges. These results reflect that the CGN / NCG motif ratio and cleavage ratio can distinguish between high-methylated CpGs and low-methylated CpGs. In addition, the CGN / NCG motif ratio and cleavage ratio can be used to predict the range of methylation levels. Each figure shows a calibration curve that can be used to determine quantitative values ​​(at least within a range) using the amount of free DNA fragments ending at positions 0 or -1.

[0277] SVM regression was also tested using a cleavage pattern of an 11 bp window (e.g., -5 to 5). The SVM model was trained using a cleavage pattern corresponding to CpGs with different methylation index ranges. For example, an SVM regression model was trained based on ranges of 0 to 5% (Group 1), 50 to 55% (Group 2), and 95 to 100% (Group 3), i.e., SVR 1. Another SVM regression model was trained for 0 to 10% (Group 1), 45 to 55% (Group 2), and 90 to 100% (Group 3), i.e., SVR 2. Another SVM regression model was trained for 0 to 20% (Group 1), 40 to 60% (Group 2), and 80 to 100% (Group 3), i.e., SVR 3.

[0278] Table 2 shows the performance of methylation status prediction for single CpG sites within different methylation index ranges using an SVM regression model based on the proximal cleavage pattern. As shown in column 2 (Pearson r), the predicted methylation score shows a good correlation with the methylation index of the tested CpG (i.e., Pearson r = 0.73, 0.77, and 0.74). For each SVM regression model, the methylation index of each predicted group is very consistent with the actual methylation index range. In some embodiments, the predicted values ​​can be expressed at (but not limited to) 99%, 95%, 90%, 85%, 80%, 75%, 70%, 65%, 60%, 55%, 50% confidence levels and other confidence intervals. Table 2. Performance of methylation status prediction using SVM based on cleavage patterns for single CpG sites within different methylation index ranges. G. Methods for Determining Methylation at Sites in a Genome

[0279] Methods are provided for determining methylation at sites in a genome using cleavage information. Methylation can be a characteristic of a sample at a location in a genome, for example, as defined relative to a reference genome. The cleavage information can be various types of cleavage patterns, as described herein. Various standardizations can be used, and cleavage information for more than one location can be used.

[0280] Figure 27 27 is a flow chart of a method 2700 for determining methylation at a site based on fragmentation according to an embodiment of the present disclosure. Various examples of the method 2700 are described above, and the method 2700 can be performed partially or completely using a computer system.

[0281] At block 2710, a plurality of free DNA molecules from a biological sample of the subject are analyzed. In any of the methods described in the present disclosure, various techniques can be used for such analysis. For example, sequencing can be used for analysis, such as high-throughput parallel sequencing, targeted sequencing, and single-molecule sequencing (e.g., using nanopore sequencing or using real-time single-molecule sequencing (e.g., from Pacific Biosciences)). Example PCR techniques include real-time PCR and digital PCR (e.g., droplet digital PCR). Analysis can include the physical steps of performing such analysis and receiving measurement data obtained from such analysis, or can include only receiving measurement data.

[0282] Analyzing the free DNA molecule can include determining the genome position corresponding to at least one end of the free DNA molecule in the reference genome.For example, any one of the various comparison techniques as understood by the technical staff can be used by one or more sequence readings of DNA molecule (for example paired readings at the end or the reading of whole molecule) and reference genome.Can be compared with some or all of the reference genome.As another example, the technology based on probe can be differentiated by DNA molecule as from a specific location, for example, by emitting a specific color for the specific probe corresponding to the specific genome location.Position determination can be for some or all of the reference genome, if for example only a part for analyzing genome.As an example, the amount of the genome analyzed can be greater than 0.01%, 0.1%, 1%, 5%, 10% or 50%.This type of analysis can be carried out for additive method described herein.

[0283] As with other methods described herein, analyzing a plurality of free DNA molecules can include measuring the size of the free DNA molecules. The measurement can be performed in various ways, such as using physical separation (e.g., electrophoresis) and / or sequencing (e.g., whole molecule sequencing or using paired end read alignment).

[0284] At block 2720, a first amount of free DNA molecules at a first position within a window around the CpG site is determined. The first position can be between positions -1 and +1 of the window. For example, the first amount can be the free DNA at the -1 position, the first amount can be the free DNA at the 0 position, or the first amount can be the free DNA at the +1 position.

[0285] The first quantity can be determined in various ways. For example, when the analysis uses sequence reads, block 2720 can determine the number of reads that align with a CpG site in the reference genome and end in CG (5' end on the left on the Watson strand) or end in GC (5' end on the right on the Crick strand). The quantity can be a count of DNA molecules. As another example (e.g., Figure 4 As depicted in , probe-based techniques can be used. For example, multiple reactions that are positive for a particular probe can be used. As yet another example, the intensity (e.g., electrical or optical) of the signal from a particular probe (e.g., from real-time PCR) can be used.

[0286] The second amount of the free DNA molecules at the second position (different from the first position) in the window around the CpG site can be optionally determined at block 2720. The second amount can be determined in a manner similar to the first amount. This amount can be an original number (for example, not standardized), because they are compared with each other. In other examples, relative frequency can be used. The example of relative frequency includes cutting ratio and cutting density, as can be determined using standardization techniques described herein. These two amounts can be used to determine relative frequency, and are used in subsequent steps. As a further example, the DNA molecules of the same number can be analyzed, the reading of the same number can be produced, or the sample of the same volume (for example, a specified volume (for example, milliliters)) can be used to carry out this type of standardization or any other standardization mentioned herein.

[0287] The second amount can be a larger amount of part.For example, it is possible to determine the total number (also referred to as depth) of the DNA molecules covering the first position (for example, CpG site). The end of at least some DNA molecules in these DNA molecules is located at different positions except the first position. Therefore, the total number covering the first position can include that the end is located at the second amount of the second position around the CpG site. The technology used to carry out standardized use of this type of total number is the second amount. Therefore, when the first amount is standardized, standardization can use the number of the free DNA molecules covering the CpG site. As another example, standardization can use the number of the free DNA molecules in the region comprising the CpG site using an end. As another example, standardization can use the average depth or median depth of the free DNA molecules in the region comprising the CpG site.

[0288] The second position can be any position within the measurement window, an example window of which is described herein. For example, if the first position is position 0, the second position can be located at +1 or -1 relative to the CpG site. As another example, the measurement window can be at least -2 to +2 relative to the CpG site, and thus be greater than 5 bases wide.

[0289] At block 2740, the classification of the methylation of the CpG site in the genome of the object is determined using the first amount. Alternatively, the second amount can also be used. Various methylation classifications of the site can be determined. For example, it can be determined whether a site is hypomethylated or hypermethylated in a sample or object. Therefore, as determined by analyzing a plurality of free DNA molecules in the sample, the classification of methylation can be whether the CpG site in the genome of the object is in a hypermethylated state (also referred to as a methylated state) or a hypomethylated state (also referred to as an unmethylated state). The methylated state can have a methylation density higher than a first critical value, and the first critical value is at least 70%. The unmethylated state can have a methylation density lower than a second critical value, and the second critical value is 30% or lower. The classification can be any methylation level, which can be qualitatively (e.g., high, medium or low, as can be defined using certain percentage critical values) or quantitatively (e.g., a specific amount, such as a percentage, which can have a given resolution / range) provided.

[0290] The classification can be determined in various ways. For example, the separation value can be determined using the first amount and the second amount. The separation value can be compared to a calibration value. The calibration value can be determined using free DNA molecules from one or more calibration samples and located at CpG sites with known classifications. The known classification can be determined using methylation correlation analysis or other analysis results using methylation correlation analysis. The calibration value can be a critical value / cutoff value, above or below which a specific classification can be determined. For example, Figure 17 In the assay, a calibration value of about 22 for a cleavage ratio at position 0 can distinguish hypermethylated sites from hypomethylated sites.

[0291] As another example of a classification, a site can be identified as having a methylation level of a specific percentage or percentage range (e.g., 5%, 10%, 15%, 20%, 25%, 30% or less). An example of a 20% range is 40% to 60%. Thus, the classification can be a quantitative value. Comparing the separation value to the calibration value can include comparing the separation value to a calibration function, as described above in Section E.

[0292] Classification can use at least two positions except the first position, for example, if the first position is position 0, then the position except the CpG site.For each position of at least two positions in the window, it is possible to determine that the end is positioned at each amount of the free DNA molecule at this position. The first amount of the free DNA molecule at the end of the first position can be compared with the each amount of the free DNA molecule at the end of the position, as a part for determining classification. This type of comparison can determine the difference between the amount at position-1 to position 0 and the difference between the amount at position 1 to position 0. For example, Figure 8 and Figure 17 Each has a different amount that increases from -1 to 0 and decreases from 0 to 1. Additionally, such use of two or more positions can include machine learning techniques, such as those described above in Sections C and D.

[0293] When using machine learning techniques, the feature vector can include the individual quantities and the first quantity and be input into the machine learning model as part of determining the classification. Any such machine learning model described herein can be trained using free DNA molecules located at CpG sites with known classifications. The first position and at least two positions can include all positions within a window of at least +5 to -5 from the CpG site. As described in Sections C.2 and D, the machine learning model can use the sequence context within the window. In one example, the machine learning model is trained for the sequence context within the window (e.g., Figure 21A ).

[0294] In another example, the feature vector may include sequence context (e.g., Figure 22 In such an example, the feature vectors may form a matrix with each row corresponding to a base in a chain (possibly two chains, a Watson chain and a Crick chain). Figure 22 As shown in , a column may include a non-zero amount at a corresponding position in a row corresponding to a base. When the feature vectors form a matrix, a CNN model may be used.

[0295] More than one CpG site can be within the measurement window (e.g., Figure 10 、 Figure 14 and Figure 15) and can be classified. In this case, the CpG site can be the first CpG site with the first classification. The window can include the second CpG site and can include at least two positions except the first CpG site (for example, if the first position is 0) and the second CpG site. The third amount of the free DNA molecules at the end located at the second CpG site can be determined. For each position of at least two positions in the window, it can be, for example, as described above, to determine the respective amounts of the free DNA molecules at the end located at the position. When using machine learning, the feature vector can include each amount, the first amount, and the third amount. The feature vector can be input into the machine learning model as a part for determining the classification of the first CpG site and determining the second classification of the second CpG site.

[0296] The methylation classification of more than one CpG site can be determined using a first amount and a second amount of a first CpG site at a cleavage site, such as described in Section II.A.3. Thus, some embodiments can use the first amount and the second amount to determine the classification of methylation of different CpG sites in the genome of a subject. The different CpG sites can be located downstream of the 5' end of the CpG site, for example, within a specified range such as 50 nt, 75 nt, or 100 nt. IV. Predicting methylation in regions based on terminal motifs

[0297] In some embodiments, the methylation of a region can be predicted. The classification of methylation can be similar to that of a site, such as a classification of hypomethylation or hypermethylation or a specific methylation level, which can be a range of values ​​(e.g., within 1%, 2%, 3%, 4% or 5%). Various techniques can be used, including machine learning. A. Size of the area to be analyzed

[0298] Region can be whole genome or genomic specific part, and it can be located on a chromosome or span multiple chromosomes.Region can be made up of the different windows that separate (that is, do not intersect) from each other.Therefore, methylation level can be for can be in multiple regions (for example, 2,3,4,5,10,20,50 etc.) on multiple chromosomes (for example, 2,3,4,5,10,20,50 etc.) determine on the whole.Following first chapter analyzes whole genome, and chapters and sections subsequently focus on genomic specific part. 1. Overall

[0299] Figure 28 The methylation index of CGN and NCG terminal motifs according to embodiments of the present disclosure is shown. Each point represents the average methylation index of each of the four motifs within the group of CGN and NCG terminal motifs. N represents A, C, G, or T. Figure 28 Similar to Figure 2B The distances are 0 and 1. CGN corresponds to Figure 7Position 0 in , and NCG corresponds to Figure 7 Position -1 in .

[0300] The figure shows that high methylation indices (median: 88%) are observed at CpG sites associated with 5'-CGN terminal motifs (i.e., CGA, CGT, CGC, CGG). In contrast, low methylation indices (median: 69%) are present at CpG sites associated with 5'-NCG terminal motifs (i.e., ACG, TCG, GCG, CCG). Cutting at methylated or unmethylated C is very different: if methylated, cutting is more likely to occur at C, while if unmethylated, cutting is less likely to occur at C. If it cuts at C, the terminal motif will be CGN. If it cuts at the previous base, the terminal motif will be NCG.

[0301] Given this difference, some embodiments can use terminal motifs to predict regional methylation density. For example, regions with more CGNs will have higher methylation density. Regions with more NCGs will have lower methylation density.

[0302] For example, one or more calibration samples can have measured methylation levels and amounts of CGN terminal motifs and / or NCG terminal motifs. The amounts of each terminal motif can be used separately or together, for example, as a ratio of the two amounts. If used separately, the amounts can be standardized, for example, using the standardization techniques described herein. The measured methylation levels (known classifications) and the parameters of one or more amounts can include calibration data points, where the parameters can be considered calibration values. Such calibrations are described in more detail in other sections, such as the Terminology section and Section III.E.

[0303] Free DNA molecules in a new biological sample can be analyzed to determine a parameter for the amount of CGN terminal motifs and / or the amount of NCG terminal motifs. This parameter can then be compared to calibration data points from one or more calibration samples, which can be accomplished using a calibration function (e.g., a regression function fit to the calibration data points of the calibration sample set). If the measured parameter is similar to the calibration value of the calibration data point, then the new sample and the previous calibration sample will have similar methylation levels. The calibration data points can be used to train a model, such as a linear model, a nonlinear model, or other similar or more complex machine learning model. 2. Select the region

[0304] based on Figure 28Based on the data in , it is assumed that the number of CGN and NCG motifs present in a specific genomic region can be used to predict regional methylation density. To this end, for example, the CGN and NCG motif frequencies and methylation densities of those plasma DNA molecules overlapping with the region near the transcription start site (i.e., ± 1000bp of TSS, referred to as the TSS region) are determined. The overlap of molecules is determined by mapping sequence reads to those regions. Each TSS region does not overlap with another TSS region. The average depth of each CpG exceeds 2. Each region has more than 100 overlapping DNA fragments.

[0305] TSS regions with a methylation density exceeding 70% were defined as hypermethylated regions (although other percentage ranges can be used), and regions with a methylation density below 30% were defined as hypomethylated regions. Regions in between were excluded but can be included in other embodiments, for example to identify partially methylated regions. More than three classifications can also be used. CGN and NCG end motif frequencies were calculated for all hypomethylated and hypermethylated regions and normalized by the expected motif frequencies from 8 healthy control samples (median paired-end sequencing reads of 381 million (IQR: 353 million to 404 million)).

[0306] The region may be a region having various sizes (eg, 100 kb, 200 kb, 500 kb, 1 Mb, 2 Mb, 3 Mb, and 5 Mb), with the 1 Mb region being merely an example. B. Example Classification Techniques

[0307] One example technique uses the amount of CGN or NCG, or the ratio of the two, possibly normalized to either. Another example technique uses the amount of each motif (e.g., each 3-mer motif). Higher k values ​​for k-mer motifs may be used. 1.CGN or NCG

[0308] In some embodiments, the amount of CGN and / or NCG can be used. This amount can be the relative frequency of the motif within the 3-mer group of the terminal motifs of the free DNA molecules being analyzed (e.g., DNA molecules from the region being analyzed (which can be the entire genome)). For example, the sum of the individual percentages of the terminal motifs CGA, CGC, CGG, and CGT in all 3-mer terminal motifs of all DNA fragments being analyzed can be determined. The same operation can be performed on NCG. Each value can be used individually or the values ​​of CGN and NCG can be used together to determine a parameter, such as a ratio. Such percentages are a type of standardization. Another example of standardization is the ratio of the amount of CGN and NCG, such as CGN / NCG, NCG / CGN, CGN / (CGN+NCG), etc. Such various ratios can be used in any embodiment of the use of the ratio of CGN to NCG disclosed herein. In some embodiments, such standardized values ​​can also be measured, for example, using expected frequencies.

[0309] In some embodiments, the expected frequency of the terminal motif can be based on the reference sequence in the region of the reference genome, such as the number of times a specific terminal motif occurs in the region of the reference genome is determined. The accurate expected frequency will depend on the sequence of the region and can be standardized, such as the size of the region can be defined as the total number of k-aggregate terminal motifs in the region. The expected frequency can provide information about whether the measured frequency is higher than the expected frequency, because some regions may have more CpG sites than other regions.

[0310] In this analysis, the expected frequency of a certain terminal motif can be determined by computer simulation to be used for adjusting the weight of the observed frequency of the corresponding terminal motif. The 3nt interval (that is, 3nt sliding window) is used to scan the region of interest with a 1nt step length. In other words, the 3nt sliding window is moved along the region of interest, offset by 1nt at each time. Therefore, a plurality of 3 polymer motifs can be generated based on the movement of the 3nt sliding window. The frequency of those 3 polymer motifs generated by computer simulation is regarded as the expected frequency of the terminal motifs relevant to the region of interest. Therefore, the expected frequency is the frequency that random cutting will occur.

[0311] In an exemplary embodiment, the observed frequency of a certain terminal motif and the ratio of expected frequency (O / E ratio) can be used for downstream analysis.In O / E ratio, O is the observed frequency (that is, standardized amount) of one group of specific one or more k-aggregates terminal motifs.Frequency can be determined via any standardization technique described herein.For example, the observed frequency can be determined as the percentage of the fragment with one in the specific k-aggregates terminal motif of this group in all k-aggregates terminal motifs (for example, 3-aggregates terminal motifs).

[0312] Figure 29The O / E ratios of CGN and NCG terminal motifs in hypermethylated and hypomethylated regions from 8 healthy control samples according to embodiments of the present disclosure are shown. As shown, the O / E ratios of CGNs in hypermethylated regions (median = 2.07) are substantially higher than those in hypomethylated regions (median = 0.99) (P value: 0.00016, Mann-Whitney U test). In contrast, the O / E ratios of NCGs in hypomethylated regions (median = 0.71) are higher than those in hypermethylated regions (median = 0.47) (P value: 0.00016, Mann-Whitney U test). These data indicate that the use of CGN and NCG motifs will allow for inference of hypomethylation or hypermethylation.

[0313] Such data can be used to train models, for example, to determine calibration values ​​(e.g., cutoff / critical values ​​for performing binary classification or values ​​for determining narrower quantitative values, as described herein). For example, when using an O / E ratio determined from CGN frequency, a value above a cutoff of about 1.7 can indicate hypermethylation, as can be determined from Figure 29 When using NCG, a value above about 0.6 can indicate hypomethylation, while a value below about 0.6 can indicate hypermethylation.

[0314] If such training data points (calibration data points) have specific values ​​of methylation levels (not just hypermethylated or hypomethylated states) in addition to calibration levels (i.e., O / E ratios), such training data points can be used to determine a calibration function (e.g., using regression) that can provide a methylation level input O / E ratio. Such use of calibration data points is described throughout this disclosure. One skilled in the art will appreciate that the applicability of such descriptions for one technique may be applicable to other techniques.

[0315] In other embodiments, a ratio of CGN / NCG can be used, and the ratio can be normalized by an expected ratio of expected frequencies. Although written as CGN / NCG, the ratio can have various forms, such as a CGN / NCG motif ratio or an NCG / CGN motif ratio or CGN / (CGN+NCG) or NCG / (CGN+NCG). Such ratios can also include ratios of functions of such values. Such variations can be used in any of the embodiments described herein. 2. Feature vector of k-mer

[0316] Some of the examples above use only the amounts of some terminal motifs, for example, only the CGN terminal motifs of all 3-mer terminal motifs (k for k-mers can use other values). This section describes the use of amounts of all k-mers. Such amounts can form a feature vector input to a machine learning model, which can, for example, determine a binary classification or provide a numerical value, such as a specific percentage (or equivalently, a decimal value). Regression can provide such numerical values.

[0317] Figure 30 Figure 2 shows the use of 3-mer terminal motifs in TSS regions to determine binary classification or regression, according to an embodiment of the present disclosure. The terminal motifs are shown in the lower left corner. As shown, 3-mer terminal motifs are used, but other values ​​of k can be used, such as 1, 2, 4, etc. For 3-mer motifs, there are 64 values ​​in the feature vector. For 4-mer motifs, there are 256 values ​​in the frequency vector that can be input into the machine learning model.

[0318] The objective was to classify the methylation density of a region. Two methods were presented. The first was a binary classification that distinguished hypermethylated regions from hypomethylated regions. Low and high were defined as a methylation density less than 30% or a methylation density greater than 70%. Regions with methylation densities between these two numbers were excluded from the analysis. A regression model was also used to quantitatively predict the exact methylation density, which could include regions with methylation densities between the cutoff values ​​used to define hypermethylated and hypomethylated regions.

[0319] Figures 31A to 31B An example training procedure according to an embodiment of the present disclosure is presented. Figure 31A Demonstrates training on a single object and labeling a single sample strategy. Regions within a single sample are trained and tested. Over 8,000 regions meeting the criteria are selected. Regions are divided into a randomly selected training set, and the remaining regions are used as a test set. There is no overlap between these two sets. Figure 31A The technology used to generate Figure 32A 、 Figure 32B 、 Figure 33A 、 Figure 33B and Figure 36 The data in .

[0320] Figure 31B Demonstrates a training strategy using multiple subjects and labeled multiple samples. Seven healthy control cases were combined to obtain over 43,000 regions for training. The model was tested using an individual with approximately 8,000 regions. Figure 31B The technology used to generate Figure 34A 、 Figure 34B and Figure 38 The data in .

[0321] SVM was implemented to predict whether the TSS region had a methylation density of more than 70% or less than 30% by using the regional 5' k-mer end motif frequency (e.g., the first three nucleotides at the 5' end of the cfDNA fragment and a total of 64 types of 3-mer motifs).

[0322] FIG. 32A to FIG. 32B and Figures 33A to 33B The binary classification results of different k-mer motifs according to embodiments of the present disclosure are shown. These results use a one-sample strategy. It is observed that the prediction accuracy increases from 1-mer motifs to 4-mer motifs. The increase from 2-mer motifs to 3-mer motifs is the most significant. The 3-mer terminal motif achieves prediction of regional methylation density from the terminal motif pattern with an AUC of approximately 0.97.

[0323] Figure 34A The performance of regional methylation status prediction using SVM based on regional terminal motif frequency or terminal motif O / E ratio according to the embodiments of the present disclosure is demonstrated. The model was trained using the terminal motif frequency or terminal motif O / E ratio of TSS regions (i.e., ±1000bp of TSS) with a methylation density greater than 70% or less than 30% from 7 healthy control samples. ROC analysis was performed on the TSS region from an independent healthy control sample to distinguish the methylated state from the unmethylated state of the genomic region. The solid line and the dotted line represent the model performance based on the terminal motif frequency and the terminal motif O / E ratio, respectively.

[0324] SVM models trained on the 5' 3-mer terminal motif frequency and the O / E ratio of the 5' 3-mer terminal motif achieved AUCs of 0.97 and 0.94, respectively. These data indicate that using the 5' terminal motif provides useful information for predicting methylation density in a specific region of interest. Note that the use of the 5' terminal motif is merely an example. In addition to or in place of any of the embodiments described herein, the 3' terminal motif can be used.

[0325] In another embodiment, support vector regression (SVR) is implemented to predict the methylation density of the TSS region based on the frequency of the regional 5' 3-mer terminal motif. SVR can perform multidimensional linear regression. The model is trained using the exact number of methylation densities and the percentage of methylation density is predicted.

[0326] Figure 34BThe performance of regional methylation density estimation using SVR based on regional end motif frequency according to embodiments of the present disclosure is demonstrated. The model was trained using regional end motif frequencies of TSS regions (i.e., ±1000 bp of the TSS) from seven healthy control samples to predict the methylation density of the TSS region from an independent healthy control sample. The scatter plot shows the correlation between the predicted methylation density for the test sample and the methylation density detected by bisulfite sequencing.

[0327] The predicted methylation density is highly correlated with the methylation density inferred by bisulfite sequencing (Pearson r: 0.93). Therefore, the data show that the 3-mer terminal motif analysis allows the methylation density of the genomic region of interest to be inferred. In another embodiment, the regional terminal motif frequency can be used to predict the methylation density of other regions, such as CpG islands (CGIs), long interspersed nuclear elements (LINEs), Alu repeats, microsatellites, tandem repeats, enhancers, transposons, transcription factor binding sites, etc. In another embodiment, the terminal motif can be related to (but not limited to) 1nt, 2nt, 3nt, 4nt, 5nt, 6nt, 7nt, 8nt, 9nt and 10nt or higher at each end of the free DNA molecule. In yet another embodiment, other machine learning models can be implemented, such as decision trees, Bayesian classifiers, support vector regression, recurrent neural networks, other neural network-based models, and other methods known to those skilled in the art.

[0328] In some embodiments, only certain terminal motifs are used, such as those corresponding to CGN or NCG, or those that include only CG (ie, including CGN and NCG, and even larger motifs, depending on the value of k). C. Standardization

[0329] Sequence context itself can play a role in predicting methylation status. Various types of normalization can be performed using the sequence context of a region. Example types of normalization can use the expected frequency of each region (as described above) or normalization of each fragment. 1. Normalized motif frequency according to region

[0330] Figure 35 ] shows the normalization using the expected frequency of the region for each motif (region normalization) according to an embodiment of the present disclosure.The normalized motif frequencies of different motifs can then be used to generate a feature vector that is input to the ML model.

[0331] The expected frequency of a given motif in a region can be calculated using a sliding window, as described above. The expected frequency can provide a measure of what frequency will be expected when randomly cutting. As shown, the observed frequency in a given region is divided by the expected frequency, and the normalized value of CC is 2, indicating that the frequency is higher than expected. The normalized frequency of TT is lower than expected. This standardization can remove the influence caused by the sequence background. In this way, only the influence of the cutting preference is maintained.

[0332] Figure 36 The results of binary classification using an SVM model with background-normalized (region-normalized) terminal motifs according to embodiments of the present disclosure are shown. Figures 32A to 33B The non-standardization shown in is even worse. For 3-mer motifs and 4-mer motifs, a decrease in prediction accuracy is observed. This decrease may be because the sequence background has some information, and through standardization, this information is actually lost. Another reason may be that for 4-mer motifs, it has 256 different types of diversity. But most regions only have ends between 100 and 200 types. Therefore, the characteristics of 4-mer motifs are relatively sparse. Larger regions can solve this type of problem because more motifs will be seen. 2. Normalized motif frequency per molecule

[0333] Normalization per molecule was also investigated.

[0334] Figure 37 The display uses the expected motif frequency in the molecule for normalization of the per-molecule background according to the embodiments of the present disclosure. Although the fragments are displayed relative to the reference genome, they are normalized using the sequence of each molecule. However, when, for example, the entire fragment is not sequenced and both ends of the fragment are mapped to the reference sequence, the sequence of the corresponding molecule can be determined using the reference sequence. The reference sequence can be used to fill in the unsequenced portions of the molecule.

[0335] The example fragments shown have a terminal motif of CGT. The observed motif appears as 1. The expected motif frequency can be determined using a 3-base sliding window on the fragment, such as described herein. For example, if the fragment is 150bp, there are 148 possibilities for cutting. If only one of those possibilities is CGT, the expected frequency is 1 / 148, thereby providing a standardized motif frequency of 148. However, if there are five examples of CGT, the expected frequency is 5 / 148, and the standardized frequency will be 29.6. In this way, the expected frequency can serve as a weight for the degree of influence of each fragment on the standardized motif frequency.

[0336] Figure 38Results for unnormalized, region-normalized, and per-molecule normalized 3-mer motifs according to embodiments of the present disclosure are shown. Results using raw 3-mer motif frequencies performed best. D. Which ends provide the greatest discrimination

[0337] Research was conducted to discover which extremities provided the best discrimination of the prediction models.

[0338] Figure 39 The difference in normalized frequencies (normalized using region) of different 2-mer terminal motifs according to embodiments of the present disclosure is shown. The Y-axis shows the normalized motif frequency. If the cleavage is random, the motif frequency is expected to be approximately one after normalization. Two outliers, CC and CG, are observed. Regardless of whether the methylation is low or high, many CC ends are present. In the high methylation region, CC ends are further increased.

[0339] For the CG, in hypomethylated regions, there is no cleavage preference, as it is close to the dashed line at 1. However, in hypermethylated regions, there is a preference for cleavage at this CG. This is consistent with other results in this paper. The CG-terminal motif also shows the greatest difference between hypomethylated and hypermethylated regions, which is consistent with the results above.

[0340] Given the difference between the normalized motif ratios (frequencies) of CC and CG, such ratios can be used to distinguish (classify) hypomethylated regions from hypermethylated regions. Such classification using calibration values ​​can be as described above, for example, for Figure 29 Using region normalization, the CG-terminal motif provided the best discrimination.

[0341] Figure 40 The differences in the normalized frequencies (region normalization) of different 3-mer terminal motifs according to embodiments of the present disclosure are shown. The CGN terminal motif showed the greatest difference, which is consistent with the results above. Similarly, motifs beginning with CC and CG were observed. It can be found that the motifs CA and CT did not show very significant differences in low and high methylation regions, but the CC motif and CG motif showed some differences, with the CG motif showing more differences than the CC motif. In addition, this difference is consistent with the fact that high methylation regions have a higher amount of CG motifs. E. Methods for Determining Methylation in a Region

[0342] Methods are provided for determining methylation of a region in a genome using cleavage information. Methylation can be a characteristic of a sample or a characteristic of a specific region in a genome, where the position of the region can be determined relative to a reference genome. The cleavage information can be various types of cleavage information, such as sequence end motifs, as described herein.

[0343] Figure 41Flowchart illustrating method 4100 for determining methylation in a region according to an embodiment of the present disclosure. As with other methods described herein, method 4100 can be performed in part or in whole using a computer system. Blocks that perform similar functionality as in other methods can be performed in a manner similar to that described for those blocks in other flowcharts. Regions can include disjoint subregions (i.e., non-contiguous) on the same chromosome or on different chromosomes.

[0344] At block 4110, a plurality of cell-free DNA molecules from a biological sample of the subject are analyzed. Block 4110 can be performed in a manner similar to block 2710, and vice versa. As with block 2710, analyzing the cell-free DNA molecules can include determining the location of the cell-free DNA molecules in a reference genome, for example, when the region is smaller than the entire genome. The plurality of cell-free DNA molecules can be located in a region of the reference genome.

[0345] Position can be determined in a variety of ways as described herein, such as by comparing sequence reads with a reference genome or using a position-based probe. Position-based probes are probes whose sequences are primarily or completely based on the sequences present in the reference genome. Therefore, the combination of position-based probes can be used to indicate the genomic position of the bound DNA. The bound DNA can be a free DNA molecule found in body fluids, such as plasma, serum, saliva, urine, peritoneal fluid, cerebrospinal fluid, sweat, seminal fluid, semen, vaginal secretions, cervical fluid, etc.

[0346] A region can be a portion of the entire reference genome or genome of an object. This portion of a genome can be 10Mb or less of a specific chromosome; other sizes of regions are mentioned herein. Therefore, multiple free DNA molecules can come from one or more specific regions of a reference genome. At least one of the one or more specific regions can correspond to a specific genotype or haplotype comprising one or more alleles. In this way, the methylation of genotypes, haplotypes, and specific chromosome copies can be determined. If an object has certain methylation levels under a genotype or haplotype, such characteristics can indicate pathological changes (e.g., via genomic imprinting, described in more detail elsewhere in the present invention). For example, certain methylation levels (e.g., specific methylation classifications) can silence genes associated with pathological changes, which may be caused by abnormal conditions / disorders of diseases (e.g., cancer, inflammation) or physiological processes (e.g., during cell differentiation or during development, or aging) or abnormal conditions / disorders associated with physiological processes. Copy number aberrations can be such pathological changes. The embodiment can also determine normal levels of physiological processes, such as monitoring the stage of pregnancy or aging of an individual, or the type or level of an individual's genomic imprinting.

[0347] The analysis can include determining the terminal sequence motif (also referred to as the terminal motif) of at least one end of the free DNA molecule. The terminal motif can be determined based on the outermost base in the sequence read. If the sequence read only includes one end rather than the entire DNA fragment / molecule (e.g., as one of the reads covering both ends), the bases corresponding to the ends of the DNA fragment are used and the bases located in the middle of the fragment are not used. The end of the free DNA molecule can have a first position located at the outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position.

[0348] At block 4120, a first amount of one or more terminal sequence motifs of the first group having a C at the first position and a G at the second position (e.g., the amount of any one of CGN or more of these motifs) is determined. The one or more terminal sequence motifs of the first group may include one or more of CGA, CGC, CGG, and CGT, where CGN represents all. In other embodiments, the first amount may be the amount of one or more terminal sequence motifs of the first group having a C at the second position and a G at the third position (e.g., NCG). In this example, the one or more terminal sequence motifs of the first group may include one or more of ACG, CCG, GCG, and TCG, where NCG represents all. Such example terminal motifs can be used in various ways as described herein to determine the amount and / or normalized parameters (e.g., frequency) for classification. Additional normalization based on the expected frequency per region or per fragment can be used.

[0349] Block 4120 can be performed in a similar manner to block 2720. For example, the first amount can be determined using sequencing or a probe-based technique (e.g., PCR). The number of sequence reads having a particular end motif can be counted. As another example (e.g., Figure 4 ), a probe-based technique can be used. For example, multiple reactions that are positive for a particular probe can be used. As another example, an intensity signal (e.g., electrical or optical) from a probe specific for a particular terminal motif can be used to determine the first amount.

[0350] At block 4130, the methylation classification of the region in the genome of the subject is determined by comparing the first amount to a calibration value. The calibration value can be determined using free DNA molecules from one or more calibration samples and located at regions with known classifications, each including one or more CpG sites. The known classification can be determined using a methylation correlation analysis or other analysis results utilizing a methylation correlation analysis.

[0351] Other quantities, such as those used for normalization, can be determined. For example, when a first quantity is determined using one or more CGN motifs, a second quantity of one or more terminal sequence motifs of a second group (e.g., NCG) having a C at the second position and a G at the third position can be determined. Determining the classification can then include normalizing the first quantity with the second quantity to obtain a normalized first quantity that is compared to a calibration value. The second group of one or more terminal sequence motifs can include all or one or more of ACG, CCG, GCG, and TCG.

[0352] Another example of a quantity is an expected quantity, as described above. The expected quantity of one or more terminal sequence motifs of the first group can be determined based on the reference sequence of the reference genome in the region. Then, determining the classification can include normalizing the first quantity with the expected quantity to obtain a standardized first quantity (e.g., O / E ratio) that is compared with the calibration value.

[0353] As another example of normalization, the total amount of sequence motifs in a plurality of free DNA molecules (e.g., regions) can be determined. Then, determining the classification can include normalizing a first amount with the total amount to obtain a normalized first amount (e.g., a motif ratio or relative frequency) that is compared to a calibration value.

[0354] As other technologies described herein, the feature vector of multiple amounts can be provided to machine learning model, such as described in above chapters and sections. Different amounts can correspond to a series of terminal motifs, such as those terminal motifs and extra terminal motifs included in the first group. Such terminal motifs included in the feature vector can be all k-aggregates or only subsets for specific k. Therefore, it is possible to determine that there is C at the first position and each amount (that is, four amounts of CGN) of each 3-aggregate terminal motif with G at the second position. The feature vector comprising each amount (it includes the first amount) can be generated, and input into the machine learning model as a part for the classification of the methylation of the region in the genome of the definite object. The machine learning model can be trained using the free DNA molecules from one or more calibration samples. Additional or alternatively, the feature vector can be included in each amount of having C at the second position and each 3-aggregate terminal motif (that is, NCG) with G at the third position. As mentioned above, the feature vector can include each amount of all 3-aggregate terminal sequence motifs. V. Predicting methylation sites in specific fragments

[0355] Some embodiments can use the terminal motif pattern associated with a single plasma DNA molecule and other features such as fragment size and sequence context to predict its methylation status. The methylation status can be fully methylated, fully unmethylated, or partially methylated. In addition to the terminal motif, various types of information can also be used to determine the methylation level of CpG sites on individual fragments. A. Input

[0356] Figure 42 An example workflow for a machine learning model for predicting methylation status at the single molecule level according to embodiments of the present disclosure is shown. A reference genome 4210 is shown, aligned with fragments 4205. For each fragment, three types of information are shown: terminal motifs, genomic (sequence) content, and size. This information can be formed into various data structures for building feature vectors. An example of a feature vector is shown in Figure 42 middle.

[0357] On the left, the terminal motif is converted into a matrix with one-hot encoding for each position. For example, the terminal motif is CGCA. The first base is C, so the first row will be one for C and the other bases will be zero. The second base is G, so the second row will be one for G. The third base is C; the fourth base is A. Therefore, the terminal motif is encoded from the 5' end based on the matrix of the one-hot encoding rule. A 4-dimensional vector is used to encode the nucleotides. For example, [1,0,0,0], [0,1,0,0], [0,0,1,0], and [0,0,0,1] represent A, C, G, and T, respectively. Therefore, for the terminal motif of 'CGCA', the matrix will be represented as Figure 42 The terminal motif may be represented in other ways, such as by assigning values ​​1 to 4 to each base type, such as A=1, C=2, G=3, and G=4.

[0358] In the middle, the use of size is shown. In this example, fragment size is determined by the sequence reading based on paired end sequencing and the comparison with reference to genome. When the whole molecule is sequenced, comparison is not needed. Standardization can be carried out, for example, by dividing by 600, because most of the free DNA fragments are less than 600bp. Therefore, fragment size can be determined by the outermost genome coordinates of the paired end readings compared and standardized (represented by feature II) by a factor of 600bp.

[0359] On the right, genome context is a matrix comprising dinucleotide patterns. The first base in the 2 polymer is on a row, and the second base is on a column. For each specific 2 polymer, the matrix element stores the count of the number of examples of this 2 polymer. Therefore, sequence context can be encoded into a matrix comprising the count of the dinucleotide type along a single plasma DNA molecule. By scanning a single plasma DNA molecule with a 2nt sliding window, the dinucleotide frequency of a single plasma DNA molecule can be determined. 2-dimensional (2D) matrices are used to express dinucleotide frequencies (represented by feature III). For example, if the frequency of "AT" dinucleotide is 2, then "2" is filled into the lattice of "A" rows and "T" columns. Other k-aggregate arrays (vectors, matrices, or high-order tensors) can be used. For trinucleotide patterns, arrays will be three-dimensional. Genome (sequence) context can also be the entire sequence. Another example of genome context is the GC percentage.

[0360] Two or more of these features can be combined (or only the terminal motifs alone) to be input into a machine learning model to predict whether the molecule is fully methylated or completely unmethylated, or partially methylated. A CNN model is trained using a combined feature matrix containing the aforementioned features (I, II, and III) to determine whether a DNA fragment is fully methylated or completely unmethylated. B. Results

[0361] Figure 43A Demonstrating the performance of single-molecule methylation status analysis using a CNN based on terminal motifs, sequence context, and fragment size according to embodiments of the present disclosure. A model was obtained that distinguished methylated from unmethylated molecules with an AUC of 0.85. The results suggest that using fragmentomic features with other features derived from a single plasma DNA molecule may be able to determine the methylation status of that plasma DNA molecule.

[0362] Figure 43B The performance of methylation status analysis of single molecules using molecules with CGN and NCG terminal motifs based on CNN models according to embodiments of the present disclosure is demonstrated. Thus, accuracy is determined by a subset of a set of DNA fragments. Interestingly, if the focus is on those molecules that begin with CGN or NCG terminal motifs, the model performance can be significantly improved (i.e., AUC of 0.85 to 0.94; P value <0.0001, DeLong test). Therefore, selective analysis of subsets of DNA molecules with certain motifs will allow for enhanced classification capabilities in inferring the methylation status of single molecules.

[0363] The CGN and NCG motifs can act as tags in DNA fragments, which can more accurately distinguish fully methylated from fully unmethylated DNA fragments at the individual molecular level. Therefore, on a per-fragment basis, embodiments can determine the methylation level of a particular fragment. C. Methods for predicting methylation of fragment sites

[0364] As described above, methods are also provided for determining the methylation status at a single fragment site using cleavage information. The cleavage information can be various types of cleavage information, as described herein, such as sequence end motifs.

[0365] Figure 44 A flow chart illustrating a method 4400 for predicting methylation of a DNA segment according to an embodiment of the present disclosure. As with other methods described herein, method 4400 can be performed in part or in whole using a computer system. Blocks that perform similar functionality as in other methods can be performed in a manner similar to that described for those blocks in other flow charts.

[0366] At block 4410, one or more sequence reads of a test cell-free DNA molecule may be received. The test cell-free DNA molecule may be generated in various ways, such as using the techniques described for blocks 2710 and 4110. The test cell-free DNA molecule may include one or more CpG sites.

[0367] At block 4420, one or more sequence reads are used to determine the sequence background of the test free DNA molecule. The example of sequence background includes the combination of bases that occur in sequence (for example, various k-mers, also referred to as K-mers). For example, sequence background can include a K-mer matrix of multiple examples of each K-mer in a specified test free DNA molecule, where K is an integer. When K equals 2, the K-mer matrix is ​​a dinucleotide matrix of multiple examples of dinucleotide pairs in a specified test free DNA molecule. As another example, sequence background can include the sequence information (that is, complete sequence) of each nucleotide from the test free DNA molecule.

[0368] At block 4430, one or more sequence reads are used to determine the first terminal motif of the first end of the test free DNA molecule. The number of bases in the first terminal motif can vary. For example, different k-polymers can be used. If the entire molecule is a sequence read, the outermost k base can be selected from either end of the sequence read. If two sequence reads are obtained from two ends, the base corresponding to the end of the DNA molecule can be identified based on the positional information obtained in the sequencing method. If a probe is used, the signal corresponding to the specific probe provides the sorting from the outermost to the center of the DNA molecule. The second terminal motif of the second end of the test free DNA molecule can also be determined.

[0369] At block 4440, method 4400 includes loading a machine learning model into the memory of a computer system, and the machine learning model is trained using a training set of training free DNA molecules with CpG sites with known methylation status. The training set can include sequence background and terminal sequence motifs for training free DNA molecules. For example, methylation-related sequencing can be used or all CpG sites can be artificially methylated or unmethylated to determine the methylation level of the training free DNA molecules. For example, whole genome amplification can make the newly synthesized DNA unmethylated, and CpG methyltransferase (M.SssI) can make the input DNA sample methylated. Training can relate to various techniques and models, including set models. For example, different subsets of training samples can be used, and different subsets can be used for verification.

[0370] At block 4450, the machine learning model can determine the methylation level of one or more CpG sites of the test free DNA molecule. The sequence background and the first end motif of the test free DNA molecule (for example, as a part of a feature vector) are input to the machine learning model. This type of feature vector can use other features. For example, the second end motif of the other end also can be input to the machine learning model. Another example feature is the size of the test free DNA molecule.

[0371] When the test free DNA molecule includes multiple CpG sites, the methylation level can be the percentage of methylation of the multiple CpG sites. As another example, the methylation level can be whether a specific CpG site is methylated.

[0372] Various formats of feature vectors can be used. For example, the first end motif can be uniquely encoded for input to a machine learning model. The format of the sequence context (or genomic context) can be a two-dimensional or higher dimensional matrix, such as based on the K value for a K-polymer.

[0373] As mentioned above, size can be input into the machine learning model as a part for the methylation level of one or more CpG sites determining the test free DNA molecule.Size can have a variety of formats, such as base number or relative to the ratio of reference size, such as the ratio of 600bp.Size can be measured in a variety of ways.For example, when a sequence reading crosses over the test free DNA molecule (for example, as the result of single molecule sequencing), the size of the test free DNA molecule can be determined by counting the base number in a sequence reading.As another example, when obtaining two sequence readings (paired ends) corresponding to the two ends of the test free DNA molecule, size can be measured by two sequence readings with reference to genome comparison.

[0374] Various types of machine learning models can be used, such as convolutional neural networks, recurrent neural networks, transformer models (encoder-decoder models), random forest classification, and other models described elsewhere in this disclosure. A machine learning model can also be a combination of models, thus an ensemble model. VI. Detection of 5hmC-enriched or -depleted regions

[0375] Many of the above discussions generally discuss methylation, rather than any particular type of methylation, because it is generally applicable. Technology can be used to detect a certain type of methylation. For example, hMe-Seal can be used to detect 5hmC methylation. Differentiate the different fragments in the regions where 5hmC is enriched (the methylation ratio from 5hmC is higher than in other regions) and 5hmC is missing (the methylation ratio from 5hmC is lower than in other regions). In this way, some embodiments can detect certain regions where the methylation ratio of a certain type is higher or lower than other regions of the genome. Based on cutting, some embodiments can determine whether a region is enriched in 5hmC or lacks 5hmC. A. Relationship between terminal motifs and regions of 5hmC enrichment and 5hmC depletion

[0376] 5-Hydroxymethylcytosine (5hmC) is an important mammalian DNA epigenetic modification that has been associated with gene regulation and cancer pathogenesis. In bisulfite sequencing, both 5mC and 5hmC are read as cytosine and are therefore indistinguishable. It is speculated that the presence of 5hmC may change the frequency of CG-terminal motifs. To test this hypothesis, 5hmC sequencing (hMe-Seal) data based on selective chemical labeling was obtained from 10 plasma DNA samples from healthy individuals (Song et al. Cell Res 2017; 27(10): 1231-1242).

[0377] hMe-Seal is a method for using β-glucosyltransferase (βGT) to selectively label 5hmC with biotin via azide-modified glucose for enrichment of DNA fragments containing 5hmC for sequencing. Such biotin can be captured by some magnetic beads that capture 5hmC fragments rather than 5mC. The higher the sequencing depth around a CpG site, the higher the amount of 5hmC present in the sample. Those higher sequencing depth regions that form a "peak" signal (higher coverage) in the entire genome are defined as "5hmC enriched regions". By using a peak recognition method (such as MACS2), 38,168 common 'peak' regions in 10 healthy subjects were identified. 4,631,823 CpG sites were analyzed and determined to be methylated in 8 healthy samples (internal data) by bisulfite sequencing. Among them, 25,102 CpGs overlapped with the peak regions of 5hmC enrichment (referred to as group I; 5hmC-enriched regions), and 3,409,329 CpGs were not present in any peak regions of the 10 healthy subjects (referred to as group II; 5hmC-depleted regions).

[0378] It was speculated that those CpG sites in group I would have relatively more 5hmC modifications, while those in group II would have relatively less 5hmC modifications. The frequencies of CG-containing terminal motifs were obtained in these selected CpGs in the same eight healthy plasma samples.

[0379] In order to compare the motif frequency between group I and group II, the expected terminal motif frequency of the motif comprising CG in these two groups is calculated from the reference genome for standardizing the observed terminal motif frequency. Standardized value is calculated by observation / expected ratio (O / E ratio). The expected frequency can be determined as described herein, for example, by using a sliding window with size k (corresponding to the k-polymer terminal motif used) to determine the respective frequency in the analyzed terminal motif. In other embodiments, only the observed frequency can be used. Subsequently, certain terminal motif frequencies between the two types of regions (5hmC enrichment and 5hmC deletion) are compared.

[0380] Figures 45A to 45B To show the CGN between 5hmC-enriched and 5hmC-depleted regions ( Figure 45A ) and NCG( Figure 45B ) motif O / E ratio box plot. Figures 45A to 45BAs shown, compared to cfDNA molecules associated with Group II CpG 4510 (absent), cfDNA molecules associated with Group I CpG 4520 (enriched) exhibited higher O / E ratios for CGA, ACG, and TCG terminal motifs. For CGC, CGG, CCG, and GCG motifs, significantly lower O / E ratios were observed in 5hmC-enriched region 4520 (i.e., Group I CpG sites). These results indicate that the presence of 5hmC will affect plasma DNA cleavage patterns.

[0381] In the above, it was shown that the O / E ratio of CG motifs is different between regions with and without 5hmC. The presence of 5hmC may change the frequency of CG motifs and the O / E ratio of CG motifs. Changes in CG motifs can be used to predict the presence of 5hmC in genomic regions.

[0382] To illustrate the predictive power, the common 'peak' regions identified above were used (n = 38,168). The length of the peak regions was then standardized to 200 bp (i.e., peak center + / - 100 bp) and the peak regions (n ​​= 24,221) containing at least one CpG site in each of the NCG backgrounds (i.e., ACG, CCG, GCG, TCG) were further selected. Those regions were referred to as regions with 5hmC in healthy controls (also referred to as 5hmC enrichment).

[0383] To define regions without 5hmC in healthy controls (also known as 5hmC depletion), the reference genome was divided into 200bp windows and windows that did not overlap with regions with 5hmC and contained at least one CpG site (n = 992,991) in each of the NCG contexts (i.e., ACG, CCG, GCG, TCG) were identified. For example, WGBS reads of plasma DNA from eight healthy control samples were pooled and classified as regions with or without 5hmC. Approximately 2,000 regions with 5hmC and 2,000 regions without 5hmC were ultimately used for downstream analysis, and these regions were covered by more than 20 CG-containing ends.

[0384] Using a dataset of 1,000 regions with 5hmC and 1,000 regions without 5hmC, an SVM model was trained using the O / E ratio of CG motifs (i.e., ACG, TCG, CCG, GCG, CGA, CGT, CGC, and CGG). The trained model was then used to analyze another independent dataset of 1,000 regions with 5hmC and 1,000 regions without 5hmC.

[0385] Figures 46A to 46BDemonstrates the performance of 5hmC prediction based on 5' 3-mer terminal motifs containing CG. Figure 46A Figure 2 is a box plot showing the probability of the presence of 5hmC between regions with 5hmC and regions without 5hmC. Figure 46B Figure 3 shows ROC analysis for distinguishing regions with 5hmC from those without. As shown, the predicted probability for regions with 5hmC is significantly higher than that for regions without 5hmC (P < 0.001). ROC analysis shows that the SVM model can achieve an AUC of approximately 0.90, distinguishing regions with 5hmC from those without 5hmC, that is, distinguishing regions with 5hmC enrichment from those without 5hmC.

[0386] In addition to the O / E ratio alone or using a feature vector comprising a CG motif (e.g., the SVM model shown above), other techniques described herein may also be used. For example, the CGN / NCG ratio may be used to select terminal motifs, such as terminal motifs with an inverse relationship. Terminal motif frequencies may also be used, possibly as feature vectors comprising CG motifs (e.g., 1 to 8 values ​​of different 3-mer terminal motifs). The feature vectors may be used in machine learning models, as described herein. Such amounts may be used directly without standardization for expected amounts. Other standardizations may be used, such as the total amount of terminal motifs of a particular group (e.g., CGN, NCG, or both) or all terminal motifs of a desired length (e.g., 3-mers or 4-mers). B. Method for detecting 5hmC-enriched and 5hmC-depleted regions using terminal motifs

[0387] Figure 47 A flow chart illustrating a method for measuring 5hmC methylation in a region of a subject's genome using free DNA molecules. As with other methods described herein, method 4700 can be performed in part or in whole using a computer system. Functional blocks similar to those described in other methods can be performed in a manner similar to that described for those blocks in other flow charts.

[0388] At block 4710, a plurality of cell-free DNA molecules from a biological sample of a subject are analyzed. Block 4710 can be performed in a manner similar to blocks 2710 and 4110. As with blocks 2710 and 4110, analyzing the cell-free DNA molecules can include determining the location of the cell-free DNA molecules in a reference genome, for example, when the region is smaller than the entire genome. The plurality of cell-free DNA molecules can be located in a region of the reference genome.

[0389] The analysis can include determining a terminal sequence motif (also referred to as a terminal motif) of at least one end of the free DNA molecule, which can be performed in a manner similar to block 4110. The terminal motif can be determined based on the outermost bases in the sequence read. If the sequence read includes only one end rather than the entire DNA fragment / molecule (e.g., as one of a paired read covering both ends), the bases corresponding to the ends of the DNA fragment are used rather than the bases located in the middle of the fragment. The ends of the free DNA molecule can have a first position located at the outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position.

[0390] At block 4720, determine the first amount of one or more terminal sequence motifs of the first group having C at the first position and G at the second position (e.g., the amount of any one of CGN or more of these motifs). The one or more terminal sequence motifs of the first group may include one or more of CGA, CGC, CGG, and CGT, where CGN represents all. In other embodiments, the first amount may be the amount of one or more terminal sequence motifs of the first group having C at the second position and G at the third position (e.g., NCG). In this example, the one or more terminal sequence motifs of the first group may include one or more of ACG, CCG, GCG, and TCG, where NCG represents all. Such example terminal motifs can be used in various ways as described herein to determine the amount and / or normalized parameters (e.g., frequency) for classification. Additional normalization based on the expected frequency per region or per fragment can be used. Block 4720 can be performed in a manner similar to block 2720 and block 4120.

[0391] At block 4730, a classification of 5hmC methylation of a region in the genome of the subject is determined by comparing the first amount to a calibration value. The calibration value can be determined using free DNA molecules from one or more calibration samples and located at CpG sites classified by a known level of 5hmC methylation. The known classification can be determined using a 5hmC correlation analysis (e.g., as described above) or other analysis results utilizing a 5hmC correlation analysis.

[0392] Other quantities, such as those used for normalization, can be determined. For example, when a first quantity is determined using one or more CGN motifs, a second quantity of one or more terminal sequence motifs of a second group (e.g., NCG) having a C at the second position and a G at the third position can be determined. Subsequently, determining the classification can include normalizing the first quantity with the second quantity to obtain a standardized first quantity that is compared to a calibration value. The second group of one or more terminal sequence motifs can include all or one of ACG, CCG, GCG, and TCG.

[0393] Another example of amount is expected amount, as described above. The expected amount of one or more terminal sequence motifs of the first group can be determined based on the reference sequence of the reference genome in the region. Subsequently, determining the classification can include standardizing the first amount with the expected amount to obtain a standardized first amount (e.g., O / E ratio) compared with the calibration value.

[0394] As another example of normalization, the total amount of sequence motifs in a plurality of free DNA molecules (e.g., in a region) can be determined. Subsequently, determining the classification can include normalizing the first amount with the total amount to obtain a standardized first amount (e.g., a motif ratio or relative frequency) that is compared to a calibration value.

[0395] As other technologies described herein, the feature vector of multiple amounts can be provided to machine learning model, such as described in above chapters and sections. Different amounts can correspond to a series of terminal motifs, such as those terminal motifs and extra terminal motifs included in the first group. Such terminal motifs included in the feature vector can be all k-aggregates or only subsets for specific k. Therefore, it is possible to determine that there is C at the first position and each amount (that is, four amounts of CGN) of each 3-aggregate terminal motif with G at the second position. The feature vector comprising each amount (it includes the first amount) can be generated, and afterwards input into the machine learning model as a part for the classification of the methylation of the region in the genome of the definite object. The free DNA molecule training machine learning model from one or more calibration samples can be used. Additional or alternatively, the feature vector can be included in each amount of each 3-aggregate terminal motif (that is, NCG) with C at the second position and G at the third position. As mentioned above, the feature vector can include each amount of all 3-aggregate terminal sequence motifs. VII. Proportional Concentration of Clinically Relevant DNA

[0396] Fragmentation in different types of tissues has also been studied. The terminal motif groups CGN and NCG and the cleavage patterns in the regions / sites with tissue-specific methylation patterns are analyzed. Clinically relevant DNA can be used to identify the DNA of any tissue type of differentially methylated regions, such as regions of high methylation, low methylation, 5hmc enrichment or 5hmc deletion of specific tissues of clinically relevant DNA. Example tissue types include fetal tissue, tumor tissue or transplanted tissue, or tissue from specific organs, such as liver tissue. Tissue-specific alleles can be used to detect tissue-specific DNA, such as measuring calibration values ​​for calibration functions, which can be used to provide proportional concentrations using fragmentome parameters. DNA fragments from specific regions / sites can be obtained via targeted technology (also referred to as enrichment), which can be carried out via amplifying certain regions or by capturing DNA from certain regions. Further details of enrichment can be found in other chapters of the present disclosure.

[0397] In the present disclosure, paired-end bisulfite sequencing (75bp×2 (i.e., paired-end sequencing), Illumina) was performed. Genotypes for maternal buffy coat and placental / chorionic villus tissue samples were obtained using microarray-based genotyping technology (HumanOmni2.5 genotyping array Illumina), and informative SNPs (i.e., where the mother is homozygous (expressed as AA genotype) and the fetus is heterozygous (expressed as AB genotype)) were identified. Fetal-specific DNA fragments are identified based on DNA fragments carrying fetal-specific alleles at specific SNP sites. In this scenario, the B allele is fetal-specific, and it is therefore inferred that DNA fragments carrying the B allele are derived from fetal tissue. The number of fetal-specific molecules (p) carrying the fetal-specific allele (B) is determined. The number of molecules (q) carrying the common allele (A) is determined. The fetal DNA fraction in all free DNA samples will be calculated by 2p / (p+q)×100%. The amount of fetal-specific alleles varied between samples (median: 198,248; range: 147,614 to 207,735). A. Tissue-specific methylation of CGN and NCG

[0398] Because we have shown that the number of DNA molecules with CGN and NCG termini correlates with the methylation status of CpGs at or near the ends of free DNA molecules, we further explored the possibility of using the relative amount of plasma DNA molecules ending with CGN and / or NCG motifs to reflect methylation patterns. Relative amount can be measured in various ways. For example, relative amount can be measured as a ratio (e.g., CGN / NCG motif ratio, or NCG / CGN motif ratio, or CGN / (CGN+NCG) or NCG / (CGN+NCG)). Comparison of CGN or NCG with other terminal motifs can also be used.

[0399] As another example, the free DNA molecules that end is CGN can be measured relative to the total number of the free DNA molecules analyzed (for example, order-checking) or the percentage of the total number of free DNA ends. Similarly, the free DNA molecules that end is NCG can also be measured relative to the total number of the free DNA molecules analyzed (for example, order-checking) or the percentage of the total number of free DNA ends. In addition, this type of standardization or any other standardization mentioned herein can be carried out by analyzing the DNA molecules of the same number, producing the reading of the same number or using the sample of same volume (for example, specifying milliliters). Therefore, if between an operation to another operation, the DNA molecules of the same number are ordered, this amount can be directly used, because effectively standardized. This type of various options for standardized can be used by any one of the methods described herein.

[0400] The results below show that the relative amounts of plasma DNA molecules ending in CGN or NCG motifs at genomic regions that exhibit differential DNA methylation patterns across different tissue types can determine the tissue of origin of the plasma DNA molecules and determine the proportional concentration of specific types of free DNA molecules. 1. Fetal-specific allele motif frequency

[0401] Maternal plasma samples were analyzed for regions that were hypermethylated and hypomethylated for the fetus. All 3-mer motifs were calculated for the selected regions, and motif frequency was determined as the amount of one type of motif (e.g., CGN or NCG) / the total amount of 3-mer motifs in the region.

[0402] Figure 48A and Figure 48B Motif frequencies for fetal-specific hypermethylated and hypomethylated regions are shown according to embodiments of the present disclosure. Depending on the methylation status, CGN and NCG terminal motif frequencies differ between DNA molecules carrying fetal-specific alleles and common alleles.

[0403] Fetal-specific and shared cfDNA from a pregnant woman in her third trimester were analyzed (sequencing depth: 270X). For the fetal-specific hypermethylated region (chr3: 187,767,589 to 187,768,381, which has nine CpG sites), the frequency of CGN-terminal motifs in fetal-specific DNA molecules was 5.2%, higher than that in DNA molecules carrying the allele shared by the fetus and the mother (1.3%). In contrast, the frequency of NCG-terminal motifs in fetal-specific DNA molecules was 1.7%, lower than that in DNA molecules carrying the allele shared by the fetus and the mother (3.2%). For the fetal-specific hypomethylated region (chr7: 25,889,274 to 25,890,315, which has nine CpG sites), a different pattern was observed compared to the hypermethylated regions. Fetal-specific DNA molecules showed a lower frequency of CGN terminal motifs but a higher frequency of NCG terminal motifs compared to corresponding DNA molecules carrying alleles shared by the fetus and mother (CGN terminal motifs: 1.7% vs. 3.7%; NCG terminal motifs: 3.3% vs. 0.5%). These results further indicate that using CGN and NCG terminal motif frequencies can provide information for predicting methylation changes across multiple genomic regions of interest.

[0404] This data demonstrates that CGNs and NCGs correlate with the amount of fetal DNA. Because the ratio of fetal DNA ends with CGNs differs from that of shared allele DNA, the amount of DNA molecules ending in CGNs can reflect the proportional concentration of fetal DNA. This is also true for DNA molecules ending in NCGs. Furthermore, because this behavior occurs in both hypermethylated and hypomethylated regions, both types of regions, CGNs and NCGs, can be used to provide four types of measurements, or only one of these types of measurements can be performed. 2. Ratio of NCG to CGN

[0405] To test the ability to determine the proportional concentration of fetal DNA, fetal-specific sites were identified and tested. Bisulfite sequencing results of buffy coat (sequencing depth: 75X haploid genome coverage) and placental tissue (sequencing depth: 94X haploid genome coverage) were used to identify placenta-specific hypermethylated (n=188,978) and hypomethylated (n=2,013,795) CpG sites. Placenta-specific hypermethylated sites were defined by those CpG sites with a methylation density greater than 70% in placental tissue but less than 30% in buffy coat samples. Placenta-specific hypomethylated sites were defined by those CpG sites with a methylation density less than 30% in placental tissue but greater than 70% in buffy coat samples. The ratio of plasma DNA molecules with CGN as the terminal motif to plasma DNA molecules with NCG as the terminal motif was determined in placenta-specific hypermethylated CpG sites and hypomethylated CpG sites, respectively. A total of 30 plasma DNA samples from pregnant women were analyzed using high-throughput parallel paired-end sequencing (median number of paired reads: 206 million; IQR: 142 million to 232 million).

[0406] Additionally, fetal-specific alleles, particularly single nucleotide polymorphisms, are identified. The number of reads with fetal-specific alleles is used to determine the proportion of fetal DNA based on the SNP.

[0407] Figures 49A to 49B Demonstrating that embodiments of the present disclosure are based on the presence of placenta-specific hypermethylated CpG sites ( Figure 49A ) and placenta-specific hypomethylated CpG sites ( Figure 49B The correlation between the CGN / NCG ratio of plasma DNA molecules and the proportion of fetal DNA was determined. Measurements were performed for each pregnant woman at different stages of pregnancy. The shape and color of the data points indicated whether the pregnancy was in the first, second, or third trimester. The amount of CGN alone or NCG alone (or either terminal motif within these groups) could also be used.

[0408] The amount of CGNs was determined using the amount of sequence reads covering hypermethylated or hypomethylated sites that had CGs at their ends. The amount of NCGs was determined using the amount of sequence reads covering hypermethylated or hypomethylated sites that had NCGs at their ends.

[0409] like Figure 49A and Figure 49BAs shown in , the CGN / NCG ratios from placenta-specific hypermethylated CpG sites and hypomethylated CpG sites were positively and negatively correlated with the proportion of fetal DNA inferred by fetal-specific SNPs, respectively (Pearson r: 0.90 and 0.86; for both correlations, P values ​​< 0.0001). The data indicate that the relative amount of plasma DNA molecules ending in CGN and NCG (e.g., CGN / NCG ratio) can be used to infer the likelihood of the tissue of origin of plasma DNA molecules and can determine the proportion of free DNA of clinically relevant DNA, such as fetal DNA, tumor DNA, or transplant DNA.

[0410] Figures 49A to 49B The line in is an example of a calibration function. For any CGN / NCG ratio determined for a new sample, this value can be input into a linear function to provide an output on the X-axis of the fetal DNA ratio. Such a calibration function can be considered as a calibration value including calibration data points, which also include corresponding proportional concentrations. The determination of proportional concentrations can include a comparison of the CGN / NCG ratio (or other features described herein) with a calibration value of a known corresponding proportional concentration. The known proportional concentration can be determined via some other techniques, such as SNP analysis. Such a calibration function only needs to be determined once, so there is no need to perform such SNP-based analysis on a new sample (which may not be applicable to all cases or take more time, analysis, and work to identify SNPs). The calibration function can be more complex, such as machine learning techniques (such as support vector regression), which will still involve comparison with calibration values. 3. Independent NCG or CGN

[0411] In addition to using the ratio of CGN / NCG, CGN and NCG can also be used independently. In one embodiment, a measurement window (e.g., 11bp) centered around each placenta-specific hypermethylated or hypomethylated CpG site can be defined as a placenta-specific hypermethylated region and a hypomethylated region. Calculate the CGN and NCG motif frequencies in the placenta-specific hypermethylated region and the hypomethylated region. Each region has a CpG site. The motif frequency (e.g., the total amount of the ends in the (CGN or NCG) / region) is determined by standardizing the amount of CGN or NCG with the total amount of the end motifs from the region.

[0412] Figures 50A to 50B Demonstrating that embodiments of the present disclosure are based on the presence of placenta-specific hypermethylated CpG sites ( Figure 50A ) and placenta-specific hypomethylated CpG sites ( Figure 50B The correlation between the frequency of NCG and CGN motifs in plasma DNA molecules and the proportion of fetal DNA. Figure 50AAs shown, the frequencies of NCG and CGN motifs derived from placenta-specific hypermethylated regions were positively and negatively correlated, respectively, with the proportion of fetal DNA inferred from fetal-specific SNPs (Pearson r: -0.84 and 0.87; P values ​​< 0.0001 for both correlations). In contrast, the frequencies of NCG and CGN motifs derived from placenta-specific hypomethylated regions were positively and negatively correlated, respectively, with the proportion of fetal DNA inferred from fetal-specific SNPs ( Figure 50B ; Pearson r: 0.87 and −0.69; P value < 0.0001 for both correlations).

[0413] The CGN motif frequency is defined as the number of terminal motifs with CGN in a region. The NCG motif frequency is defined as the number of terminal motifs with NCG in a region. Such normalization can be performed in a variety of ways. For example, as described above, the total number of free DNA molecules analyzed (e.g., sequenced) or the total number of free DNA ends can be used, rather than limiting the normalization factor to DNA within a measurement window surrounding only CpG sites.

[0414] Each of these figures shows a calibration function that can be used according to embodiments of the present disclosure (e.g., as described herein). For example, a new value for the frequency of CGN motifs in a hypermethylated region can be compared to a calibration value (e.g., by inputting the value into Figure 50A The linear calibration function is shown in the figure on the right. 4. CGN / NCG of hypomethylated and hypermethylated regions in ML models

[0415] In another embodiment, the CGN / NCG ratios derived from placenta-specific hypermethylated CpG sites and hypomethylated CpG sites, respectively, can be integrated together as two independent features for machine learning. In other embodiments, a single amount (e.g., CGN or NCG) can be used as an independent feature. Examples of such features are described above.

[0416] Such features can be input to a machine learning model that can determine the proportional concentration of clinically relevant DNA. For example, a support vector regression (SVR) model can be used, for example, which can receive two (e.g., any change in the CGN / NCG ratio of hypermethylated and hypomethylated CpG sites) to four (e.g., the amount of CGN and NCG at hypermethylated and hypomethylated CpG sites) features. In other embodiments, other algorithms can be used, including but not limited to linear / nonlinear regression, decision tree, Lasso regression, random forest, elastic network, Bayesian network, support vector machine, convolutional neural network, recurrent neural network, logistic regression, etc. For example, the SVR model can be used to complete the determination of the proportion of fetal DNA in the maternal plasma DNA of a pregnant woman.

[0417] Figure 51 The correlation of the fetal DNA proportion (x-axis) inferred by the SNP-based method with the CGN / NCG ratio (y-axis) using hypermethylated and hypomethylated CpG sites from placenta-specific according to embodiments of the present disclosure is shown. In this example, the SVR model was used.

[0418] like Figure 51 As shown in Figure 2, the fetal DNA fraction estimated by the CGN / NCG ratio from pregnancy samples correlated with the fetal DNA fraction inferred by the SNP-based method (Pearson r: 0.95; P value <0.0001). This correlation held for all three trimesters. The data suggest that combined analysis of CGN / NCG ratios from regions exhibiting tissue-specific hypomethylation and hypermethylation will enhance analysis of tissues from which plasma DNA molecules are derived, such as the determination of the fraction of cfDNA from placental tissue.

[0419] Figure 51 The calibration data points shown in are examples of training data (eg, calibration data points) used to determine the SVR calibration function. Such training data is from calibration samples, as described herein for other examples. Figure 51 Demonstrates the accuracy of the SVR calibration function. 5. Examples of other organizational types

[0420] Tissue types other than fetus / placenta can be analyzed using loci with tissue-specific methylation status. For example, loci that are hypermethylated or hypomethylated in liver cells, blood cells, or colon cells can be used to determine the percentage (ratio concentration / proportion) of DNA fragments in samples from each tissue (e.g., plasma or serum). To illustrate the utility of such measurements for different tissue types, the relative changes in different tissue types from different types of subjects (transplantation and pregnancy) were analyzed relative to healthy controls.

[0421] As shown above (e.g., Figures 49A to 49B ), the CGN / NCG motif ratio of placenta-specific hypermethylated and hypomethylated CpG is highly correlated with the fetal DNA ratio in maternal plasma DNA. In addition to determining the proportional concentration, the terminal position of a given molecule at a tissue-specific locus (e.g., hypermethylated or hypomethylated in a given tissue) can be used to identify the DNA fragments most likely to be from a specific tissue. When a DNA fragment has a CGN motif (e.g., indicating the possibility of methylation) at a hypermethylated locus in a specific tissue, the possibility that the given DNA fragment is from a specific tissue is higher. In this way, the terminal motif can be used to enrich the sample for DNA fragments from a specific tissue type.

[0422] Figure 52 Schematic diagram showing the CGN / NCG motif ratio analysis based on tissue-specific methylated CpGs in blood and liver. Figure 52 Shown are cfDNA fragments ending around CpG site 5205, which is hypermethylated in blood cells; CpG site 5215, which is hypermethylated in liver; CpG site 5225, which is hypomethylated in blood cells; and CpG site 5235, which is hypomethylated in liver. Schematic showing how CGN / NCG motif ratio analysis can be performed on a panel of tissue-specific methylated CpGs across different tissue types.

[0423] Tissue-specific hypermethylated CpGs were defined as CpGs with a methylation density greater than 70% in the target tissue (e.g., liver) and less than 30% in the remaining tissues analyzed. Tissue-specific hypomethylated CpGs were defined as CpGs with a methylation density less than 30% in the target tissue and greater than 70% in the remaining tissues analyzed. The CGN / NCG motif ratio of tissue-specific CpGs was calculated for each group in plasma DNA samples.

[0424] For tissue-specific hypermethylated CpGs, the target tissue contributes more DNA molecules that are cleaved at 5' relative to the 'C' site in the CpG context to the plasma DNA pool, thus producing a higher CGN / NCG motif ratio. Other tissues have the opposite preference at those CpG sites. Therefore, the CGN / NCG ratio of molecules derived from those tissue-specific hypermethylated CpG sites is directly proportional to the amount of cfDNA contributed by the target tissue. In contrast, for tissue-specific hypomethylated CpGs, the target tissue contributes fewer plasma DNA molecules that are cleaved at 5' relative to the 'C' site in the CpG context, thus producing a lower CGN / NCG motif ratio. The CGN / NCG ratio of molecules derived from those tissue-specific hypomethylated CpG sites is inversely proportional to the amount of cfDNA contributed by the target tissue.

[0425] Since the concentration of DNA from blood cells in plasma is higher than that from the liver, the CGN / NCG ratio at blood cell-specific hypermethylated loci is higher than that at liver-specific hypermethylated loci. This relationship is shown at the bottom of the panel for hypermethylated CpGs. Conversely, the CGN / NCG ratio at blood-specific hypomethylated loci is lower than that at liver-specific hypomethylated loci. This relationship is shown at the bottom of the panel for hypomethylated CpGs.

[0426] For example, bisulfite sequencing of buffy coat, liver tissue, colon tissue, and placenta identified 7,467, 3,756, 7,726, and 1,000 hypermethylated CpG sites specific to blood cells, liver, colon, and placenta, respectively, as well as 40,788, 19,619, 93,510, and 631,734 hypomethylated CpG sites specific to the corresponding tissues. The ratios of plasma DNA molecules ending with CGN-terminal motifs to those ending with NCG-terminal motifs were determined for these tissue-specific hypermethylated and hypomethylated CpG sites, respectively. High-throughput parallel paired-end sequencing was used to analyze plasma DNA samples from a healthy control (391 million paired-end reads), a liver transplant patient (35 million paired-end reads), and a pregnant woman (238 million paired-end reads).

[0427] Figures 53A to 53B Demonstrating distinct tissue-specific hypermethylation in plasma DNA from healthy controls, liver transplant cases, and pregnant cases ( Figure 53A ) and hypomethylation ( Figure 53B ) CGN / NCG motif ratios of CpGs. For each tissue type (blood, colon, liver, and placenta), the motif ratios are presented for three different subjects (healthy controls, liver transplants, and pregnant women). The relative values ​​of the motif ratios indicate increases and decreases in the DNA proportion of each tissue across different subjects.

[0428] like Figure 53A As shown in Figure 2, in healthy controls, the CGN / NCG motif ratio for hypermethylated CpG sites originating from blood cells is highest, followed by the liver, and lowest from the placenta. This is due to the fact that liver transplant recipients contribute more liver DNA, thus reducing the contribution from blood cells. Similarly, pregnant women have free fetal DNA, thus reducing the blood contribution.

[0429] for Figure 53BAt tissue-specific hypomethylated CpG sites in the plasma, the CGN / NCG motif ratios appear in reverse order. A higher CGN / NCG motif ratio at tissue-specific hypermethylated CpGs will reflect a higher methylation density in plasma DNA at those tissue-specific CpG sites, indicating a higher contribution from that tissue. A lower CGN / NCG motif ratio at tissue-specific hypomethylated CpGs will reflect a lower methylation density in plasma DNA at those tissue-specific CpG sites, indicating a higher contribution from that tissue.

[0430] These results are consistent with the concept that blood cells are the main source of free DNA in plasma (Sun et al. Proc Natl Acad Sci USA. 2015; 112: E5503-5512). In liver transplant cases where the proportion of liver cfDNA is approximately 40% (based on the method described by SNPs), the CGN / NCG motif ratios derived from liver-specific hypermethylated and hypomethylated CpG sites, respectively, showed substantial increases and decreases compared to plasma DNA from healthy controls. The liver contribution is approximately 10% in healthy controls.

[0431] In pregnancies where the proportion of fetal cfDNA was approximately 39% (SNP-based approach), the CGN / NCG motif ratios derived from placenta-specific hypermethylated and hypomethylated CpG sites showed substantial increases and decreases, respectively, compared to plasma DNA from healthy controls. The colon contribution remained approximately the same across subjects.

[0432] The data show that the relative amount of plasma DNA molecules with ends of CGN and NCG (e.g., CGN / NCG ratio) can be used to infer the tissue of the source of plasma DNA molecules from multiple tissues. Therefore, the relative change (increase or decrease) of different tissue DNA ratios can be determined using motif ratios. For example, the baseline motif ratio of a healthy control can be used to determine whether a specific tissue DNA ratio has changed significantly, such as by comparing the new motif ratio of the object of the motif ratio for tissue-specific loci. The specific proportion contribution of any specific tissue type can be determined using each calibration function for a specific tissue type. B. Specific methylation patterns of cleavage patterns

[0433] In addition to or in lieu of using terminal motifs (e.g., CGN and NGC), a cleavage pattern can be used. The use of a cleavage pattern can be similar to that described above. One or more quantities (e.g., cleavage ratios / densities) can be determined. Such quantities can be used together with one or more known proportional concentrations as described herein and calibration values ​​(e.g., as a calibration function) to form calibration data points. Thus, a cleavage amount at a given position or a cleavage pattern at two or more positions can be used. 1. Cleavage patterns at hypermethylated and hypomethylated sites

[0434] Figures 54A to 54B Demonstrating that fetal-specific methylation is reflected in the cleavage pattern of cfDNA fragmentation according to embodiments of the present disclosure. In this figure, black line 5405 represents the cleavage pattern of fetal-specific DNA, while gray line 5410 represents the cleavage pattern of fetal common DNA. Figure 54A The cleavage patterns of fetal-specific DNA and common DNA molecules derived from placenta-specific hypermethylated CpG sites are shown. The x-axis represents the position relative to the CpG site of interest. The y-axis represents the average cleavage ratio. Figure 54B Cleavage patterns of fetal-specific and common DNA molecules are displayed, originating from placenta-specific hypomethylated CpG sites.

[0435] The cleavage patterns of fetal-specific and common plasma DNA molecules from one pregnancy cfDNA sample were determined based on placenta-specific hypermethylated (n=188,978) and hypomethylated (n=2,013,795) CpG sites, respectively. Bisulfite sequencing results of buffy coat (sequencing depth: 75X) and placental tissue (sequencing depth: 94X) were used to identify placenta-specific hypermethylated and hypomethylated CpG sites. Placenta-specific hypermethylated sites were defined by those CpG sites with a methylation density greater than 70% in placenta tissue but less than 30% in buffy coat samples. Placenta-specific hypomethylated sites were defined by those CpG sites with a methylation density less than 30% in placenta tissue but greater than 70% in buffy coat samples.

[0436] For placenta-specific hypermethylated CpG sites, fetal-specific fragments show a higher cleavage ratio at the CpG site than the common fragments (mainly of hematopoietic origin). In placenta-specific hypomethylated CpG sites, fetal-specific fragments show a lower cleavage ratio at the CpG site than the common fragments. These results indicate that tissue-specific cfDNA methylation differences can be reflected in the cfDNA fragmentation pattern. Methylation patterns can be used to infer the DNA contribution of different tissues / organs to the plasma DNA pool (Sun et al. Proc Natl Acad Sci USA. 2015; 112; E5503-5512). Therefore, it is speculated that the cleavage pattern will be suitable for inferring the DNA contribution of different tissues to the plasma DNA pool, such as the proportion of placental DNA in the plasma DNA of pregnant women. 2. SVR Implementation Methods

[0437] Using pregnancy samples as an example, we determined cleavage patterns from placenta-specific hypermethylated and hypomethylated CpG sites. Using these cleavage patterns from 30 pregnancy samples (median paired-end sequencing reads: 206 million (IQR: 142 million to 232 million)), we trained an SVR model to predict fetal DNA contribution based on a leave-one-out strategy. In other embodiments, SVR can use the amount at position 0 (e.g., cleavage ratio), which can be supplemented by the amounts at other positions.

[0438] Figure 55 and Figure 56 The present invention demonstrates the marker selection, data collection and training of an SVR model for determining the proportion of fetal DNA in a cell-free sample (e.g., plasma) according to an embodiment of the present disclosure. Bisulfite sequencing data from buffy coat and placenta are used. The marker criteria are the same as those mentioned in the previous section.

[0439] For data collection, individual samples were sequenced and the cleavage pattern for each CpG site was calculated. For hypermethylated CpGs, a merged cleavage pattern was calculated as the average cleavage ratio for each position. For a position, all values ​​from individual sites were averaged to produce a single value. The same procedure was performed for hypomethylated sites.

[0440] For each patient, there are two vectors: one from hypermethylated CG sites and one from hypomethylated CG sites. Each vector contains 11 values ​​reflecting the average cleavage ratio at each position. Other values ​​can be used for different widths of the cleavage pattern, as described herein, such as widths of 4, 5, 6, 7, 8, 9, and 10. For each training sample, the fetal DNA proportion is determined using the fetal-specific allele.

[0441] The SVR model was trained using a leave-one-out strategy, which involved training the model on 29 samples and then predicting the fetal DNA ratio for one sample. Subsequently, the model was trained on a different set of 29 samples and tested on another sample until all 30 samples were in the test set and the fetal DNA ratio had been predicted. The predicted fetal DNA ratio was compared with the fetal DNA ratio predicted based on SNP analysis.

[0442] Figure 57The correlation between the fetal DNA proportion inferred by the cleavage pattern and the SNP-based method according to the embodiments of the present disclosure is shown. The fetal DNA proportion estimated from the pregnancy sample by the cleavage pattern is highly correlated with the fetal DNA proportion inferred by the fetal-specific SNP (Pearson r: 0.97; P value <0.0001). In various embodiments, the cleavage pattern can be used to predict the DNA proportion of other tissues, including but not limited to liver, brain, T cells, B cells, neutrophils, muscle, heart, placenta, ovary, breast, testis, etc.

[0443] The tissue DNA ratio can be a clinically relevant DNA ratio, such as a tumor DNA ratio. 3. Cleavage patterns at individual sites

[0444] Above example combines the measurement across CpG sites and uses the cutting pattern of gained subsequently.For example, the ratio of the amount of the DNA molecule that end is positioned at 0 position can be compared with the amount (or both sum) of the DNA molecule that end is positioned at-1 or 1 position.Also can use or replace other positions, as described above.As another example, cutting pattern can be input to machine learning model.

[0445] In other embodiments, the independent amount of each CpG site can be determined and compared with each calibration value respectively.Therefore, if analyzing N CpG sites, then N amount can be determined, each amount is corresponding to the amount of the DNA fragmentation at each CpG site.In addition, the independent cutting pattern of each CpG site (or some site subsets) can be determined, and these independent cutting sites can be used, for example, input to the machine learning model.Each cutting pattern can be input to the machine learning model to determine the proportional concentration of this specific CpG site.This type of proportional concentration can be averaged (comprising the weighted average based on sequencing depth).

[0446] In another example, all cleavage patterns can be input to the machine learning model simultaneously. In this way, more features will be input to the machine learning model. The size of such a feature matrix can be the number of CpG sites multiplied by the number of positions in the measurement window.

[0447] In various examples, separate cleavage ratios can be determined for each cleavage pattern for different subsets of CpG sites, and the individual ratio concentrations can be averaged. C. Determining the proportional concentration of DNA from the first tissue using tissue-specific methylation sites

[0448] The proportional concentration of free DNA molecules (cell-free DNA fragments) from a particular tissue type (e.g., clinically relevant DNA) can be determined in a variety of ways. For example, the amount of DNA fragments in a new sample that end at positions 0 or -1 of one or more CpG sites and calibration data (e.g., from a calibration sample with a known or otherwise measured proportional concentration) can be used to determine the proportional concentration in the new sample.

[0449] Figure 58 Flowchart illustrating method 5800 for determining the proportional concentration of clinically relevant DNA according to embodiments of the present disclosure. A biological sample contains cell-free DNA. As with other methods described herein, method 5800 can be performed in part or in whole using a computer system. Blocks that perform functionality similar to that described in other methods can be performed in a manner similar to that described for those blocks in other flowcharts.

[0450] At block 5810, method 5800 includes analyzing a plurality of cell-free DNA molecules from a biological sample of the subject. Analyzing the cell-free DNA molecules may include determining a genomic position in a reference genome corresponding to at least one end of the cell-free DNA molecule. Such analysis may be performed on each of the plurality of cell-free DNA molecules in block 5810 using any of the techniques described herein, such as those described herein with respect to other methods, such as in block 2710. In some embodiments, the analysis of the cell-free DNA molecules may include determining a terminal motif at at least one end of the cell-free DNA molecule, wherein the end of the cell-free DNA molecule has a first position that is located at an outermost position, a second position that is immediately adjacent to the first position, and a third position that is immediately adjacent to the second position.

[0451] The analysis can be performed in various ways (e.g., as described herein) and can include physical steps controlled by a computer. As an example, the analysis can be performed by sequencing or PCR (e.g., droplet digital PCR) or mass spectrometry or electrophoresis.

[0452] At block 5820, method 5800 includes identifying the first group of CpG sites that all have the first differential methylation for the first tissue type in the reference genome. As an example, the first differential methylation can be that all the CpG sites of the first group are low methylation or that all the CpG sites of the first group are high methylation. Other examples are that all the CpG sites of the first group are 5hmC enriched or that all the CpG sites of the first group are 5hmC deleted. In some embodiments, for example, when the amount of each type is used as a separate feature of a machine learning model, two types of CpG sites can be used. In such embodiments, another type of CpG site of the second group can be identified (for example, if the first group is low methylation, the second group is high methylation) and another amount is determined for the site of the second group.

[0453] The first group of CpG sites may include one or more CpG sites, such as one CpG site or a plurality of CpG sites.

[0454] At block 5830, method 5800 includes determining a first amount of free DNA molecules that terminate at any of the 0 or -1 positions of any of the CpG sites in the first group. Such an example corresponds to using a CGN motif at the 0 position. The first amount of CpG sites that terminate at the 0 or -1 position can be further filtered to use only a subset of CGN motifs, such as CGA, CGT, CGC, or CGG, or a subset of these four terminal motifs. Such combinations can also be used for other techniques described in other sections herein. Other amounts can be determined, for example, at another position around the CpG site, for example to determine a CGN / NCG ratio or cleavage pattern.

[0455] In another embodiment or as part of determining different features of the model, an NCG terminal motif corresponding to the -1 position can also be used. For example, a first quantity of one or more terminal sequence motifs of the first group has a C at the second position and a G at the third position. The terminal motifs of the first group can include one or more of the 3-mer motifs ACG, CCG, GCG, and TCG.

[0456] Thus, a first quantity of one or more terminal sequence motifs of the first group can be determined. The one or more sequence motifs of the first group can have a C at the first position and a G at the second position. As another example, the one or more sequence motifs of the first group can have a C at the second position and a G at the third position.

[0457] The first amount can be standardized before being compared with the calibration value.Standardization can be carried out in various ways.For example, standardization can use the number (for example, as above regional standardization) of the free DNA molecules in the region that terminal is positioned at comprising CpG site.As another example, standardization can use the number (for example, using sequencing depth, as described herein) of the free DNA molecules covering CpG site.As another example, standardization can use the average depth or the median depth of the free DNA molecules in the region that comprises CpG site.

[0458] When analyzing more than a CpG site, instead of the single amount of each CpG site, each amount at each CpG site can be determined (for example, as described in Section B.3 above). Therefore, determining the first amount can include determining each amount of the free DNA molecules at the 0 or -1 position of each site in the CpG site of the first group. Subsequently, determining the proportional concentration of the DNA from the first tissue type can include comparing each of the individual amounts with each calibration value. Such individual amounts can be used in the eigenvector of the model. Therefore, by inputting each of the individual amounts into the machine learning model as a part for the eigenvector, each of the individual amounts is compared with each calibration value. The whole cutting pattern (see above B.3) of each site can be used. Therefore, for each CpG site, each amount of the free DNA molecules at the position can be determined, for example, by determining that the end is located at the position (for example, each position about at least two positions in the window around the CpG site). Each amount of at least two positions and CpG sites can be included in the eigenvector.

[0459] At block 5840, method 5800 includes determining a proportional concentration of DNA from a first tissue type in the biological sample by comparing the first amount to a calibration value. The calibration value can be determined from one or more calibration samples having a known proportional concentration of DNA from the first tissue type. The proportional concentration of DNA can have various resolutions, such as above or below a certain percentage or within a range. Such a range can have various resolutions, such as 5%, 10%, 15%, 20%, 25%, and 30% resolution or lower. As an example, a 5% resolution can correspond to 35% to 40%.

[0460] In various embodiments, such techniques can be used to determine the DNA contribution of various tissues, including but not limited to liver, brain, T cells, B cells, neutrophils, muscle, heart, placenta, ovary, breast, testis, tumor tissue (e.g., hepatocellular carcinoma, lung cancer, colon cancer, ovarian cancer), diseased tissue (e.g., tissue involved in inflammation (e.g., liver tissue involved in hepatitis, peritumoral tissue, or other) and tissue with increased cell death (e.g., ischemic or necrotic tissue or tissue involved in traumatic injury), and tissue infected with one or more infectious agents (e.g., viruses or bacteria).

[0461] As mentioned above, other amounts can be determined. For example, if the first amount uses the 0 position, it is possible to determine that the end is located in the first position within the window around any one of the CpG sites of the first group, such as the second amount of the free DNA molecules at the -1 position corresponding to NCG. Subsequently, determining the proportional concentration can include standardizing the first amount with the second amount to obtain a standardized first amount (e.g., CGN / NGC) compared with a calibration value. As an example, the first position can be located at +1 or -1 away from one of the one or more CpG sites. As another example, the window can be at least -2 to +2 away from the CpG site, or at least -4 to +4 away from the CpG site.

[0462] In some embodiments, cutting pattern can be determined in window such as described above.Cutting pattern can belong to at least two positions (such as first position) except one in the CpG site in the window.For each position in at least two positions in the window, can determine that end is positioned at each amount (comprising the second amount) of the free DNA molecule of this position.As a part for determining proportional concentration, the first amount of the free DNA molecule that end is positioned at 0 or-1 position of the CpG site can be compared with each amount of the free DNA molecule of this position with end.For example, can determine the independent ratio of each amount in each amount and the first amount.

[0463] In other embodiments, the comparison of each amount can be carried out as a part of a machine learning model (such as support vector regression).For example, a feature vector can include each amount and the first amount.The feature vector can be subsequently input into the machine learning model to determine the proportional concentration.The machine learning model can use the free DNA molecule training from one or more calibration (training) samples, and the sample has the DNA proportional concentration from the first tissue type known (such as measurement).Training can be carried out using the various techniques that those skilled in the art will understand.

[0464] When one or more CpG sites of a first group for a first tissue type in a reference genome are all hypomethylated, one or more CpG sites of a second group that are all hypermethylated (e.g., as shown in FIG. Figure 47 ). A first feature vector can be generated for the first group, and a second feature vector can be generated for the second group. Both feature vectors can be input into a machine learning model as part of determining proportional concentration. In any of these example techniques using a window, the at least two positions can include all positions within the window. As an example, the window can be at least +2 to -2 from a CpG site, or at least +4 to -4 from a CpG site.

[0465] As described in above sections A, multiple tissue types can be analyzed to determine proportional concentration or its variation.In such examples, one or more CpG sites of the second group of all hypomethylated or all hypermethylated reference genomes for the second tissue type can be identified.The second amount of the free DNA molecules in the window around any site in the CpG sites of the second group can be determined. Subsequently, the variation of the DNA proportional concentration from the second tissue type in the biological sample can be determined by comparing the second amount and another calibration value (such as the baseline in a healthy control). Therefore, another calibration value can be determined from one or more other calibration samples. D. Determining the Ratio of Concentrations of Multiple Tissues

[0466] As mentioned above, some embodiments can perform deconvolution to determine the proportional concentration of multiple tissues simultaneously. In addition, such techniques do not require identification of sites that are hypermethylated or hypomethylated in a particular tissue type. Cleavage pattern and / or terminal motif analysis can be used to estimate the proportional concentration of DNA in different tissue types.

[0467] The reference matrix A can include values ​​for a set of sites (e.g., CpG sites and surrounding sites), for example, when a cleavage pattern is used. Terminal motif values ​​can also be used. Each column in the reference matrix can correspond to a different tissue type or a different template sample (e.g., with a specified ratio of different tissues). Measurements (a vector of measurements) of the values ​​b in a cell-free mixture (e.g., plasma, urine, etc.) can be used together with A to determine the contribution percentage x of each tissue type for each template sample. This can be done by solving Ax=b.

[0468] When the columns in A correspond to different tissue types, the contribution vector x provides the percentage of each tissue type. In such embodiments, the reference values ​​in the reference matrix from tissue cells can be prepared using various DNA nucleases (e.g., DNASE1L3, DNASE1, DFFB, etc.) to digest the nuclei from those corresponding tissues. For example, DNSE1L3-based digestion can be performed on the nuclei from liver tissue. The sequencing data of the digested DNA is analyzed to obtain reference values ​​(e.g., cleavage patterns and / or terminal motifs) at multiple sites to obtain tissue-specific patterns of fragment omics features for comparison with the plasma fragment omics features of cfDNA.

[0469] In another example, a reference can be generated by analyzing tissue-specific cfDNA molecules. For example, a liver transplant case can be used to identify liver-specific DNA (e.g., using tissue-specific alleles) and use liver-specific sequence reads to determine tissue-specific patterns. Similar techniques can be used for other tissues.

[0470] When using template samples, each template sample can correspond to a set of percentages that total 100%. For example, a liver transplant subject can be used as a template, with a specified percentage for each tissue in a set of tissues. Other templates can be lesions of each organ, for example, with a template for each tissue type whose proportion is to be determined. Healthy subjects can also be used. To determine the contribution percentage of each tissue, the template's contribution is combined with the percentage of that tissue type for that template sample, which is performed for each template sample. VIII. Lesion Detection

[0471] In addition to determining the ratio concentration of methylated and clinically relevant DNA, the cleavage pattern and terminal motif groups (CGN and NCG) are used to detect diseases (e.g., cancer). For terminal motifs, global analysis is performed or the amount at a specific CpG site (e.g., a site that is hypermethylated in a specific tissue type and / or hypomethylated in a specific tissue type) is analyzed.

[0472] Specific CpG sites (e.g., sites that are hypermethylated in specific tissue types and / or hypomethylated in specific tissue types) and cleavage patterns at CpG sites that are not differentially methylated among tissues but have different cleavage patterns for different grades of lesions are also used. Different types of lesions (e.g., different types of cancer) can be determined using informative sites that are specific for different types of lesions (e.g., different types of cancer). Using cleavage information around CpG sites (e.g., terminal motifs and cleavage patterns) yields unexpectedly accurate results relative to using other terminal motifs (e.g., CCA or all 3-mer terminal motifs).

[0473] Viruses and human DNA are used in different technologies. Example lesions include cancer and autoimmune diseases (e.g., systemic lupus erythematosus (SLE)). The classification of lesions can be the presence or absence of lesions. In other examples, the classification can be a classification of the severity of the lesion, such as described herein, such as the size of one or more tumors or the surface area of ​​one or more tumors. Such severity can be determined based on the degree to which a parameter is above a critical value / cutoff value or relative to a calibration degree from a subject known to have a specific severity (e.g., cancer stage). A. Use of CGN and / or NGC terminal motifs

[0474] When the whole genome is determined to be methylated, cancer is often hypomethylated. This global hypomethylation is described in PCT publication WO2014 / 043763, which is incorporated herein by reference in its entirety for all purposes. Figure 28The fragmentation that shows CpG site place reflects methylation.Following analysis shows that pathological changes (such as cancer) can be determined using the fragmentation such as in full genome or on the repeated part of genome.This type of analysis can also be carried out for genome all or most part (such as at least 10Mb, 20Mb, 50Mb, 100Mb, 500Mb, 1Gb or more).This part of genome does not need to be continuous, for example 50Mb can be separated in five 10Mb regions that do not intersect (i.e., discontinuous). 1. CGN / NCG ratio

[0475] The genome-wide CGN / NCG ratio can be used to reflect the genome-wide methylation level and is used for disease detection. The global methylation density was analyzed in six healthy controls and eight hepatocellular carcinoma (HCC) patients, with a median of 424 million paired-end bisulfite sequencing reads (IQR: 352 million to 519 million).

[0476] Figures 59A to 59C Figure 4 shows genome-wide methylation density between healthy controls and HCC cases according to embodiments of the present disclosure. Figure 59A ), genome-wide CGN / NCG ratio ( Figure 59B ) and the CGN / NCG ratio in the Alu region ( Figure 59C ). Figure 59A As shown, HCC cases indicated significantly lower methylation levels compared to healthy controls (P value < 0.001, Wilcoxon test). The genome-wide CGN / NCG ratio of each patient was analyzed and showed that the genome-wide CGN / NCG ratio in the HCC group was significantly reduced compared to the control group ( Figure 59B ; P value < 0.001, Wilcoxon rank-sum test).

[0477] In addition to whole-genome analysis, it is assumed that changes in the CGN / NCG ratio across Alu regions or other repetitive components can be applied to reflect corresponding changes in methylation. More than 90% of methylated CpG sites in the human genome appear in DNA repetitive components, such as Alu and LINE-1 (Zheng et al. Nucleic Acids Research. 2017; 45: 8697-8711). Alu regions can effectively represent the whole genome. Other repetitive regions can also be used. As an example, Figure 59C As shown in , the CGN / NCG ratio of the Alu region in HCC cases showed a significant decrease compared to controls (P value <0.001, Wilcoxon rank sum test). These results indicate that the use of genome-wide CGN / NCG ratios enables determination of global genome-wide methylation and detection of cancer.

[0478] Therefore, global and repeat-specific fragmentation patterns specific for CGNs and NCGs (e.g., as ratios or individually) can provide accurate disease detection. Discrimination between healthy and HCC was unexpected using fragmentation values, without the need for methylation-related analysis. 2.CGN or NCG

[0479] The terminal motifs of CGN and NGC can also be used separately. The frequencies of CGN and NCG motifs are calculated across the genome and Alu regions.

[0480] Figures 60A to 60D Shown are box plots illustrating the differences in CGN and NCG motif frequencies from the whole genome and Alu regions between healthy controls and HCC cases according to embodiments of the present disclosure. Figures 60A to 60D As shown in the results, the frequencies of CGNs from the whole genome and Alu regions were significantly decreased in the HCC group compared with the control group (P value = 0.0047, Wilcoxon rank sum test). In addition, the frequencies of NCGs from the whole genome and Alu regions were significantly increased in the HCC group compared with the control group (whole genome: P value = 0.05; Alu region: P value = 0.0019, Wilcoxon rank sum test).

[0481] Using the ratio CGN / NCG (or variants thereof) allows for better discrimination, but individual amounts of these terminal motifs can also be used. 3. Multi-cancer analysis

[0482] The above examples are for HCC, but other cancers can also be detected. Because ALU has a better discriminatory effect, the CGN / NCG ratio in the ALU region is used to analyze different cancer types. A decrease in the CGN / NCG ratio in the Alu region can facilitate the detection of various cancers (i.e., pan-cancer detection). To this end, six healthy controls and eight patients with HCC, 10 patients with head and neck squamous cell carcinoma (HNSCC), 10 patients with colorectal cancer (CRC), 10 patients with lung cancer (LC), and 10 patients with NPC were analyzed, with a median of 87 million paired-end sequencing reads (IQR: 73 million to 127 million).

[0483] Figure 61 The Alu region CGN / NCG ratios from healthy controls, HCC, HNSCC, CRC, LC, and NPC according to embodiments of the present disclosure are shown. The P value between the two groups was calculated by using the Wilcoxon rank sum test. Figure 61As shown, the CGN / NCG ratio of Alu regions from cancer subjects (including but not limited to HCC, HNSCC, CRC, LC, and NPC) was significantly lower compared to control subjects (P values: HCC = 0.00067; HNSCC = 0.042; CRC = 0.0047; LC = 0.003; NPC = 0.042, Wilcoxon rank sum test). Therefore, the CGN / NCG ratio can serve as a biomarker for pan-cancer analysis. A more obvious decrease was seen for CRC, LC, and HCC. 4. PCR Analysis

[0484] PCR can be used for any of the techniques described herein, such as those described above in Section 1. Such PCR analysis can be used to determine the overall CGN / NCG ratio for cancer diagnosis, as well as to measure the methylation and proportional concentration of DNA from specific tissue types. Thus, digital or single-molecule PCR analysis can be used to determine the CGN / NCG ratio by utilizing repetitive regions, such as Alu regions. In various examples, digital PCR can be a droplet digital PCR analysis, or PCR based on other types of distribution (e.g., beads or surfaces (e.g., on glass or in a flow trough) for performing the reaction). In another embodiment, digital PCR can be implemented using, for example, a microfluidic chip from Fluidigm.

[0485] Figure 62 Example techniques and example structures of probes for determining NCG and CGN motifs in regions of interest using droplet PCR according to embodiments of the present disclosure are shown. For example, Figure 62 A probe 6210 is shown that targets a DNA template sequence 6220 that is ligated to an adaptor 6230. PCR can use a probe that covers the point at which the adaptor 6230 is ligated to the DNA template sequence 6220.

[0486] In the example shown, different probes are used for CGN and NCG. In this way, NCG can be distinguished from CGN. As an example, for the connecting portion, the probe can have a size of about 15bp. The probe has a portion with a size of, for example, 7 to 8bp that hybridizes with the adaptor region, while another portion of the probe with a size of, for example, 7 to 8bp hybridizes with the targeted Alu sequence. Those skilled in the art will understand that other sizes for the hybridization portion can be used. Observe the most CG probes and the most NCG probes for potential experimental designs, that is, the sequence with the most occurrence of 7 to 8bp of the probe in the Alu region.

[0487] Figures 63A to 63BThe top 10 CG probes and top 10 NCG probes for potential experimental designs according to embodiments of the present disclosure are shown. Different combinations were analyzed to see which probes provided the best results. Different combinations used one from CGN and one from NCG to determine the ratio.

[0488] Using the CG probe "CGCCTGT" and the NCG probe "TCGCTTGA" as examples, the number of cfDNA fragments ending in "CGCCTGT" and "TCGCTTGA" in healthy controls and cancer patients (determined using sequencing results) was used to perform a computer simulation estimate of the potential performance of ddPCR. The CGN / NCG ratio in those targeted Alu regions that should be read by ddPCR analysis will be reflected as the amount of fragments ending in "CGCCTGT" divided by the amount of fragments ending in "TCGCTTGA" in the sequencing results. The computer simulation estimate uses sequencing data to identify sequences and then selects corresponding sequences as they would be detected using PCR probes.

[0489] Figure 64 The present invention shows an in silico analysis of the CGN / NCG ratio in healthy controls and subjects with HCC, HNSCC, CRC, LC or NPC by ddPCR analysis using probes containing the Alu-related sequences "CGCCTGT" and "TCGCTTGA" according to embodiments of the present disclosure. The P value between the two groups was calculated by the Wilcoxon rank sum test. Figure 64 As shown, the CGN / NCG ratios from cancer subjects were significantly lower compared to control subjects (P values: HCC = 0.0027; HNSCC = 0.0047; CRC = 0.0005; LC = 0.001; NPC = 0.001, Wilcoxon rank sum test). This distinction worked well for all cancers, or better for some cancers than using 3-mer sequences for CGN / NCG, as shown above.

[0490] Figures 65A to 65E

[00145] ROC analysis is presented for differentiating patients with HCC, HNSCC, CRC, LC, or NPC from healthy controls using CGN / NCG ratios obtained by in silico ddPCR analysis, according to embodiments of the present disclosure. Figure 64The same two sequences used in the previous study were used. The CGN / NCG ratio inferred from the targeted Alu region, which can be analyzed by ddPCR, was able to distinguish HCC (AUC: 0.96), HNSCC (AUC: 0.92), CRC (AUC: 0.98), LC (AUC: 0.97), and NPC (AUC: 0.97) from healthy controls. Using these specific terminal motifs yielded a surprisingly good ROC of 0.872 for HCC, compared to previous work using all 3-mer motifs.

[0491] Probes of different sizes and / or complexes can be used for cancer detection. Probes targeting other regions of interest, such as LINE-1, can be used for cancer detection and monitoring. In other examples, mass spectrometry (e.g., using a time-of-flight mass spectrometer) can be used to detect reaction products comprising DNA molecules.

[0492] A further example can be to use qPCR (real-time PCR) to detect CGN and NCG frequencies from a specific region.

[0493] Figure 66 An example method is presented to identify NCG and CGN motifs in a region of interest using qPCR. Figure 4 The method described in Figure 4 The description of can be applied to this technology. As shown, free DNA molecules can undergo DNA end pairing, A tailing and adapter ligation as optional steps. A pair of PCR primers can be designed in the following manner: one primer (e.g., common forward primer 6645) can bind to the common adapter region 6615 and another primer (e.g., region-specific reverse primer 6650) binds to the specific region of interest. The DNA molecule will be amplified by a pair of PCR primers and hydrolyze a fluorescent probe specific to a certain terminal motif (e.g., CGN or NCG terminal motif) to emit a fluorescent signal, thereby being able to detect the presence of a specific motif and quantify the specific motif.

[0494] Figures 67A to 67C Examples for designing primers and probes are shown. The amplification range (i.e., between two primers) can be a repetitive region, including (but not limited to) LINEs (long interspersed modules) and SINs (short interspersed modules). For illustration purposes, a reverse primer (i.e., CCACTGCACTCCAGCCTG) targeting the consensus sequence of an Alu module (a type of SINE) is designed. Figure 67A Together with the forward primer targeting the added common adaptor, this pair of primers can be used to amplify the Alu region retaining the terminal motif at the 5' end of the cfDNA. The two probe hybridization sites (i.e., CGCCTGT and TCGCTTGA) are selected from the Alu consensus sequence of interest (i.e., 200 bp before the reverse primer) ( Figure 67A Together with the 12 nt sequence from the common adaptor, a CGN probe (ie, CTCTTCCGATCTCGCCTGT) and an NCG probe (ie, CTCTTCCGATCT TCGCTTGA) were designed ( Figure 67B ).

[0495] To test the specificity of these probes, possible hybridization sites were determined genome-wide based on sequencing reads of cfDNA. Figure 67C As shown, the majority of probe hybridization sites are located within the Alu region, with only approximately 0.3% off-target. Thus, these two probes can provide good coverage and provide good results as explained below.

[0496] This PCR-based method was tested in a mechanically fragmented gDNA sample, a healthy control cfDNA sample, and an early-stage HCC cfDNA sample. For qPCR testing, the relative fold change in the CGN / NCG motif ratio was calculated based on the CT value determined by qPCR analysis.

[0497] Figures 68A to 68B Results from qPCR (68A) and ddPCR (68B) are shown. gDNA represents a mechanically fragmented gDNA sample and was used as a control in the qPCR analysis. The fold change in the CGN / NCG motif ratio relative to the mechanically fragmented gDNA sample was based on the expression of the control (i.e., fragmented gDNA sample) (CT CGN对照 ) and test samples (CT CGN测试 ) of CGN probes and controls (CT NCG对照 ) and test samples (CT NCG测试 ) was calculated as follows.

[0498] For ddPCR testing, the CGN / NCG motif ratio was calculated as the amount of CGN probe-positive droplets divided by the amount of NCG probe-positive droplets.

[0499] As gDNA samples are mechanically sheared to 200bp, the terminal motifs present in the gDNA samples tend to be random (i.e., the CGN / NCG motif ratio is close to 1). In healthy control cfDNA samples, because Alu regions are generally highly methylated, there are more CGN terminal motifs than NCG terminal motifs (i.e., the CGN / NCG motif ratio is >1). In fact, in both qPCR and ddPCR test results, healthy control cfDNA samples showed a substantially higher CGN / NCG motif ratio than mechanically fragmented gDNA. In addition, in both qPCR and ddPCR test results, the CGN / NCG motif ratio from HCC cfDNA samples was lower than that from healthy control cfDNA, reflecting the reduced methylation level (i.e., hypomethylation) in HCC samples.

[0500] Tumor-derived DNA is shorter than background DNA, which is primarily hematopoietic in origin. The amplicon size corresponding to the CGN probe is longer than the amplicon size corresponding to the NCG probe because the CGN probe is 86nt away from the reverse primer, while the NCG probe is only 45nt away from the reverse primer. Therefore, relative to the CGN probe in cancer patients, the amplicon corresponding to the NCG probe will increase because the probability of both the NCG probe and the reverse primer being present on the same fragment is higher, that is, because the distance is shorter. The result is an additional reduction in the CGN / NCG motif ratio due to size characteristics. Therefore, targeting the CGN / NCG motif ratio in the Alu region can synergistically enhance cancer signals associated with hypomethylation and size reduction of tumor-derived DNA molecules, which will result in higher performance in cancer detection, monitoring, and prognosis. 5. Terminal motifs of fragments of specific sizes

[0501] In some embodiments, the methylation-associated fragmentation analysis described in the present disclosure can be performed using long-read sequencing technologies, including but not limited to nanopore sequencing (e.g., Oxford Nanopore Technologies) and single-molecule real-time sequencing (e.g., Pacific Biosciences). Plasma DNA samples from 15 healthy subjects were sequenced using single-molecule real-time sequencing (Pacific Biosciences) (median read count: 970,633; IQR: 407,695 to 1,399,709). Based on bisulfite sequencing of plasma DNA samples from 8 healthy subjects, putatively hypermethylated (4,631,823) and putatively hypomethylated (307,831) CpGs in plasma were identified. Putatively hypermethylated CpGs were defined as CpGs with a methylation density greater than 70% in all cases. Putatively hypomethylated CpGs were defined as CpGs with a methylation density less than 30% in all cases. The CGN / NCG motif ratios of putatively hypermethylated and putatively hypomethylated CpGs in plasma were analyzed by combining all single-molecule real-time sequencing reads from these healthy subjects.

[0502] Figures 69A to 69B The CGN / NCG motif ratios in putatively hypermethylated and putatively hypomethylated CpGs across different size ranges are shown for those single-molecule real-time sequencing reads pooled from healthy subjects. Figure 69A As shown in Figure 2, the CGN / NCG motif ratio of putatively hypermethylated CpGs decreases with increasing cfDNA fragment size. Figure 69B As shown, the CGN / NCG motif ratio of putative hypomethylated CpGs decreases with increasing cfDNA fragment size in the size range <= 1000 bp, while this value increases with increasing cfDNA fragment size in the size range > 1000 bp.

[0503] Figures 69A to 69B The results showed that cleavage preferences differed across different size ranges. For long molecules, the CGN / NCG motif ratios for hypermethylated and hypomethylated CpG sites were closer than for sizes below 200. Therefore, we investigated whether analyzing DNA fragments within a certain size range could improve accuracy.

[0504] To illustrate the usefulness of considering fragment size during methylation-related fragmentation analysis, for illustrative purposes, single-molecule real-time sequencing (Pacific Biosciences) was used to sequence plasma DNA from 15 healthy subjects, 13 chronic HBV-infected patients (HBV carriers), and 45 HCC patients. In one embodiment, the overall CGN / NCG motif ratio in healthy subjects, HBV, and HCC cases was calculated using short (<=200bp) and long (>=1000bp) plasma cfDNA molecules, respectively. Other values ​​of the short molecule cutoff value other than 200bp can be used, such as 150bp, 175bp, 225bp, and 250bp. In addition to 1000bp, other values ​​of the long molecule cutoff value can be used, such as 6000bp, 700bp, 800bp, 900bp, 1,110bp, 1,500bp, and 2,000bp.

[0505] Figures 70A to 70C Figure 70A shows the overall CGN / NCG motif ratios of short cfDNA fragments (<=200 bp; 70A) and long cfDNA fragments (>=1000 bp; 70B) in healthy subjects (healthy), HBV carriers (HBV), and HCC cases, as well as the difference in CGN / NCG motif ratios between short cfDNA fragments (<=200 bp) and long cfDNA fragments (>=1000 bp) (70C). Due to the overall hypomethylation of HCC, the overall CGN / NCG motif ratio of HCC cases in short DNA fragments was significantly lower than that of healthy subjects and HBV subjects ( Figure 70A ). Meanwhile, for long DNA fragments, HCC patients showed a higher overall CGN / NCG motif ratio compared with healthy subjects and HBV subjects ( Figure 70B ).

[0506] In some embodiments, a first amount (or ratio thereof) of short DNA fragments having CGN or NCG and a second amount (or ratio thereof) of long DNA fragments having CGN or NCG can be used together. The use of a separation value (e.g., a difference or ratio) between the two ratios can provide improved accuracy, for example, because the short DNA fragments and the long DNA fragments vary in opposite directions. As another example, the first amount and the second amount can be used as input to a machine learning model (possibly four amounts of CGN and NCG for both short DNA fragments and long DNA fragments), such as a decision tree, SVM, regression, or neural network.

[0507] Therefore, the motif ratios of cfDNA fragments with different lengths can be combined by, but not limited to, calculating the difference or ratio between the CGN / NCG motif ratios of short cfDNA fragments and long cfDNA fragments. Figure 70C, the difference in the CGN / NCG motif ratio between short and long cfDNA fragments was significantly lower in HCC subjects compared with non-HCC subjects.

[0508] Figures 71A to 71B Demonstrate the performance of HCC detection based on CGN / NCG motif ratio and size information. Figure 71A We present ROC analysis (combined analysis) for distinguishing HCC from non-HCC (healthy and HBV subjects) using the overall CGN / NCG motif ratio of short cfDNA fragments (<=200bp) and long cfDNA fragments (>=1000bp) and the difference in CGN / NCG motif ratio between short cfDNA fragments (<=200bp) and long cfDNA fragments (>=1000bp). Figure 71B We present ROC analysis (combined analysis) for distinguishing HCC from HBV using the overall CGN / NCG motif ratio of short cfDNA fragments (<=200bp) and long cfDNA fragments (>=1000bp) and the difference in CGN / NCG motif ratio between short cfDNA fragments (<=200bp) and long cfDNA fragments (>=1000bp).

[0509] The use of the CGN / NCG motif ratio measure derived from short DNA yielded an AUC of 0.89 for distinguishing HCC cases from non-HCC cases (healthy subjects and HBV subjects; Figure 71A ), and an AUC of 0.79 for distinguishing HCC from HBV carriers ( Figure 71B The use of the CGN / NCG motif ratio metric derived from long DNA resulted in an AUC of 0.72 for distinguishing HCC cases from non-HCC cases (healthy subjects and HBV subjects; Figure 71A ), and an AUC of 0.79 for distinguishing HCC from HBV subjects ( Figure 71B ).

[0510] An AUC of 0.95 could be achieved for distinguishing HCC cases from non-HCC cases (healthy subjects and HBV carriers; Figure 71A ) and an AUC of 0.90 for differentiating HCC from HBV subjects ( Figure 71B). Such performance is better than the corresponding AUC value of using short cfDNA fragments or long cfDNA fragments alone. In another embodiment, short cfDNA fragments can be defined as cfDNA fragments with a length less than (but not limited to) 50bp, 60bp, 70bp, 80bp, 90bp, 100bp, 200bp, 300bp, 400bp, 500bp, 600bp, 1000bp, etc., and long cfDNA fragments can be defined as cfDNA fragments with a length greater than (but not limited to) 300bp, 400bp, 500bp, 600bp, 700bp, 800bp, 900bp, 1000bp, 2000bp, 3000bp, 4000bp, 5000bp, etc. In another embodiment, cfDNA fragments with different sizes can be obtained from Illumina sequencing platforms, long read sequencing technologies or other platforms. Quality can also be used as a measure of size instead of length. 6. Methods using CGN and / or NGC terminal motifs

[0511] Lesions can be determined in various ways. For example, the amount of DNA fragments whose ends are located at CpG sites (e.g., CGN terminal motifs or subsets thereof) in a new sample. This amount can be determined using a group of DNA fragments within a specific size range. Classification can be determined using various models, such as a comparison of an amount with a critical value, a cutoff value, or a reference value, or a machine learning model. Such models can be trained using reference / calibration samples with known or measured classifications (e.g., via the use of different techniques). As with the training of other models, different accuracy targets can be selected, such as a desired trade-off between sensitivity and specificity. Such training can be performed for other techniques described herein.

[0512] Figure 72 A flowchart illustrating method 7200 for determining the grade of a lesion according to an embodiment of the present disclosure is provided. The biological sample includes cell-free DNA. As with other methods described herein, method 7200 can be performed partially or entirely using a computer system. Blocks that perform similar functionality to other methods can be performed in a manner similar to that described for those blocks in other flowcharts.

[0513] At block 7210, method 7200 includes analyzing multiple free DNA molecules from a biological sample to obtain sequence readings. The sequence reading can include terminal sequences corresponding to the ends of the multiple free DNA molecules. Block 7210 can be performed in a manner similar to block 2710 and other analysis steps described herein. In some embodiments, for example, when analyzing a region smaller than the entire genome, analyzing the free DNA molecules can include determining the position of the free DNA molecules in the reference genome. Such regions can be chromosomes. Multiple free DNA molecules can be located in regions of the reference genome. The position can be determined in various ways as described herein, for example, by comparing sequence readings with a reference genome or using position-based probes.

[0514] At block 7220, method 7200 includes, for each molecule in a plurality of free DNA molecules, determining the sequence motif of each end in one or more ends of the free DNA molecules.Block 7220 can be implemented via any one of the techniques described herein, for example, via sequencing or probe-based technology (including PCR) to determine the end motif.Block 7220 can be performed in a manner similar to block 4120.The end motif can be determined according to the outermost base in the sequence reading.If the sequence reading only includes an end rather than the entire DNA fragment / molecule (for example, as one of the paired readings covering both ends), the base corresponding to the DNA fragment end is used instead of the base in the middle of the fragment.The end of the free DNA molecule can have the first position, the second position next to the first position, and the third position next to the second position, which are located at the outermost position.

[0515] At block 7230, method 7200 includes determining a first amount of one or more terminal sequence motifs of a first group having a C at a first position and a G at a second position (e.g., an amount of CGN or any one of more of these motifs). For example, the first group of one or more terminal sequence motifs may include one or more of CGA, CGC, CGG, and CGT, where CGN indicates any nucleotide at the third position. The first amount can be normalized in various ways.

[0516] In other embodiments, the first amount can be the amount of one or more terminal sequence motifs (e.g., NCG) of the first group having a C at the second position and a G at the third position. In this example, the one or more terminal sequence motifs of the first group can include one or more of ACG, CCG, GCG, and TCG, where NCG represents all. Such example terminal motifs can be used in various ways as described herein to determine the amount and / or normalization parameters for classification. Block 7230 can be performed in a manner similar to block 2730 and similar steps described herein.

[0517] Other quantities, such as other terminal motifs, such as NCG, can be determined and used to determine the CGN / NCG ratio. Thus, when the first quantity uses CGN, a second quantity of a second group of terminal sequence motifs having a C at the second position and a G at the third position can be determined, the terminal sequence motif corresponding to NCG or any subset thereof. The classification can then include normalizing the first quantity with the second quantity to obtain a standardized first quantity that is compared with a reference value. As another example, the total amount of the sequence motif (e.g., in one or more specific regions) can be determined and used for standardization. The standardized first quantity (e.g., motif ratio or relative frequency) can again be compared with a reference value.

[0518] As another example of normalization, the total amount of sequence motifs in a plurality of free DNA molecules (e.g., in a region) can be determined. Subsequently, determining the classification can include normalizing a first amount with the total amount to obtain a standardized first amount (e.g., a motif ratio or relative frequency) that is compared with a reference value.

[0519] At block 7240, method 7200 includes determining a classification of the grade of the lesion of the subject based on a comparison of the first quantity to a reference value. The reference value can be determined as part of a training model, for example, using reference samples with known classifications. The comparison to the reference value can be performed in various ways, such as a direct comparison to the reference value, where values ​​above or below the reference value indicate different classifications. As another example, the comparison can be performed via a machine learning model.

[0520] As mentioned above, a plurality of free DNA molecules can be from one or more specific regions of the reference genome. The lesion can be in these one or more specific regions. For example, at least one of the one or more specific regions corresponds to a specific genotype or haplotype comprising one or more alleles. If the object has a certain amount of fragmentation of certain terminal motifs (such as CGN and NCG) under the genotype or haplotype, then such characteristics can indicate the grade of the lesion caused by the abnormal condition / disorder of a disease (such as cancer or inflammation) or a physiological process or the abnormal condition / disorder related to a physiological process (such as via aging or developmental process or genomic imprinting, described in more detail elsewhere in this disclosure). Copy number aberrations can be such lesions. The embodiment can also determine the normal level of a physiological process, such as monitoring the stage of pregnancy, aging, or the type or level of an individual's genomic imprinting.

[0521] As other technologies described herein, the feature vectors of multiple quantities can be provided to the machine learning model, such as described in the above sections. Different quantities can correspond to a series of terminal motifs, such as those terminal motifs and extra terminal motifs included in the first group. Such terminal motifs included in the feature vector can be all k-aggregates or only subsets for a specific k. Therefore, it is possible to determine the individual quantities (that is, four quantities) of each 3-aggregate terminal motif with C at the first position and G at the second position. A feature vector including each quantity (which includes the first quantity) can be generated and subsequently input into the machine learning model as a part of the classification of the lesion grade (such as the abnormal level of a physiological process) of the object. Additionally or alternatively, the feature vector can include the individual quantities of each 3-aggregate terminal motif with C at the second position and G at the third position.

[0522] Analysis of free DNA molecules can include measuring the size of the free DNA molecules. A first quantity of a first group of free DNA molecules within a first size range can then be determined, such as described in Section 5. As an example, the first size range can correspond to sizes smaller than a size cutoff value or to sizes larger than the first size cutoff value.

[0523] When the first range is greater than the first size cutoff, a second amount of a second group of free DNA molecules within a second different size range can be determined, for example corresponding to a size less than the second size cutoff. The second amount can be the amount of one or more terminal motifs of CGN or NCG as described above. The classification can then be determined using the first amount and the second amount. When CGN is used for long fragments and short fragments, a third amount of NCG long DNA fragments and a fourth amount of NCG short DNA fragments can be used, for example defined using the first size cutoff and the second size cutoff. As an example, the classification can use a ratio of the first amount to the second amount, can use a difference between the first amount and the second amount, or can use a machine learning model that receives the first amount and the second amount as separate inputs. B. Tissue-specific hypomethylated or hypermethylated regions / sites

[0524] In addition to analyzing CGN and NCG (for example, overall or on most of the genome), certain sites can also be analyzed. For example, the hypermethylation or hypomethylation sites of a specific tissue can be analyzed. Such analysis can determine whether a specific type of cancer (for example, tissue origin / root) exists. As an example, the size of a region can be at least 2bp, 10bp, 100bp, 200bp, 500bp, 1kb, 2kb, 3kb, 4kb, 5kb, 10kb, 50kb, 100kb, 500kb, 1Mb, etc. DNA fragments from specific regions / sites can be obtained via targeted technology (also referred to as enrichment), which can be carried out via amplifying certain regions or by capturing DNA from certain regions. Further details of enrichment can be found in other chapters and sections of this disclosure.

[0525] Analyze samples from HCC patients. Identify differentially methylated regions between the liver and neutrophils (the most common white blood cell), which represent the background of blood cell DNA. Each region is 500bp long. In this way, liver-specific hypermethylated regions and liver-specific hypomethylated regions compared with neutrophils are obtained. Depending on the size of the region, the differentially methylated region can include one or more CpG sites.

[0526] For hypermethylated liver regions, if enzyme (e.g., DNASE1L3) activity is comparable between disease and control groups, the amount of CGN-terminal motifs should increase in HCC patients and the amount of NCG-terminal motifs should decrease. Therefore, the CGN / NCG ratio should increase. However, if enzyme activity is downregulated, this ratio may decrease. For hypomethylated liver regions, the amount of CGN should decrease in HCC patients and the amount of NCG should increase. Therefore, the CGN / NCG ratio should decrease.

[0527] When enzyme activity (e.g., DNASE1L3) is comparable between disease and control groups, HCC subjects have a higher hepatocyte-to-plasma DNA ratio than healthy subjects. In hypermethylated liver regions, CGN is expected to increase and NCG is expected to decrease relative to healthy subjects. In hypomethylated liver regions, CGN is expected to decrease and NCG is expected to increase. This will result in an increase in the CGN / NCG ratio in hypermethylated regions, but a decrease in the CGN / NCG ratio in hypomethylated regions.

[0528] In addition, the cleavage pattern or other methylation-related cfDNA fragment omics features from liver or HCC-specific hypomethylated regions can be used for HCC detection. And the methylation-related cfDNA fragment omics features from other tissue-specific hypomethylated or hypermethylated regions can be used to detect other cancers. For example, the methylation-related cfDNA fragment omics features from colon-specific, breast-specific or prostate-specific hypomethylated or hypermethylated regions can be used for colon cancer, breast cancer or prostate cancer diagnosis, respectively. Such methylation-related cfDNA fragment omics features from specific regions can be inferred based on targeted sequencing results. Targeted sequencing can be based on (but not limited to) probe-based hybridization, amplicon-based sequencing, immunoprecipitation followed by sequencing, etc. 1. CGN and / or NCG

[0529] The frequency of motifs derived from tissue-specific hypomethylated or hypermethylated ends is used to detect and monitor diseases such as cancer. Eight healthy controls, 17 patients with chronic hepatitis B (HBV), and 34 patients with hepatocellular carcinoma (HCC) were analyzed, with a median of 483 million paired-end sequencing reads (IQR: 395 million to 549 million).

[0530] Figure 73A and Figure 73B The performance of HCC detection based on CGN / NCG motif ratios from liver-specific hypermethylated and hypomethylated regions according to embodiments of the present disclosure is demonstrated. Figure 73A Figure 2 is a box plot showing the CGN / NCG motif ratios in liver-specific hypermethylated and hypomethylated regions from healthy controls (CTR), chronic hepatitis B patients (HBV), and hepatocellular carcinoma (HCC) patients. Figure 73B ROC analysis for discriminating patients with HCC from those without using the CGN / NCG motif ratios from liver-specific hypomethylated and hypermethylated regions, respectively.

[0531] Comparing HCC subjects with non-HCC subjects, a significantly higher CGN / NCG motif ratio (i.e., CGN motif frequency / NCG motif frequency) was observed across liver-specific hypomethylated regions (P value < 0.001, Kruskal-Wallis test), while no significant difference was observed in liver-specific hypermethylated regions. The CGN / NCG motif ratio from liver-specific hypomethylated regions can serve as a biomarker for HCC, allowing detection of HCC patients with an AUC of 0.85 (P value = 0.003, DeLong test).

[0532] These results indicate that the CGN / NCG motif ratio associated with liver-specific hypomethylated CpG sites exhibits better ability to distinguish HCC patients from non-HCC subjects than liver-specific hypermethylated CpG sites. This enhancement in cancer detection is because selective analysis of the CGN / NCG motif ratio from hypomethylated regions can synergize the characteristic cfDNA cleavage caused by DNA methylation with downregulated DNASE activity (e.g., both epigenetic changes and DNASE activity changes induce the CGN / NCG motif ratio). 2. Use vectors of individual 3-mer or higher motifs

[0533] Since liver-specific hypomethylated regions can contain more information, all 3-mer motifs in this region are further used to train the SVM model. Thus, the feature vector has 64 values ​​for the amount of each 3-mer present in the hypomethylated region. Other machine learning models, such as those described herein, can be used.

[0534] Figure 74A and Figure 74B The performance of HCC diagnosis using SVM based on the frequency of 3-mer terminal motifs from liver-specific hypomethylated regions according to embodiments of the present disclosure is demonstrated. Figure 74A Figure 2 is a box plot showing the probability of HCC determined by the frequency of 3-mer terminal motifs of those plasma DNA molecules associated with liver-specific hypomethylated regions (CTR: healthy controls; HBV: chronic hepatitis B patients; HCC: HCC patients). Figure 74B ROC analysis for differentiating patients with HCC from those without HCC using the frequency of 3-mer terminal motifs from liver-specific hypomethylated regions. The frequency of 3-mer terminal motifs can be determined as the amount of a specific terminal motif divided by the total number of DNA fragments or terminal motifs detected or used.

[0535] To further maximize the utility of the terminal motifs from liver-specific hypomethylated regions, a support vector machine (SVM) model was constructed using the 5' 3-mer terminal motifs from liver-specific hypomethylated regions for HCC detection. The model achieved an area under the corrected curve (AUC) of approximately 0.91 for distinguishing HCC from non-HCC.

[0536] Therefore, the frequency of terminal motifs from tissue-specific hypomethylation or hypermethylation can be used to detect and monitor diseases, such as cancer. For example, from liver-specific hypomethylated regions, all 5' 3-mer terminal motifs are used to build an SVM model for HCC detection. However, fewer than all 64 3-mer terminal motifs can be used. For example, only eight motifs including CG can be used. These eight 3-mer motifs unexpectedly improve accuracy compared to using all 64 terminal motifs. These eight motifs are specifically CGA, CGT, CGC, CGC, ACG, TCG, CCG, and GCG. Other subsets of terminal motifs, such as subsets that do not include CG, can also be used.

[0537] The potential for HCC detection was further tested by using CG-containing 3-mer terminal motifs (a total of 8 types of terminal motifs) instead of using all 64 types of 3-mer terminal motifs. Based on the results of bisulfite sequencing of buffy coat and HCC tissues, 116,245 and 802,393 HCC-specific hypermethylated and hypomethylated CpGs were identified, respectively.

[0538] In various embodiments, any one or more of the CGN and / or NCG amounts that can be normalized (e.g., as frequencies) can be used. Frequencies can be determined in various ways. For example, normalization can be performed as performed when all 64 termini are used. As another example, normalization can be performed within a subset of terminal motifs, such as by normalizing by the number of terminal motifs within the subset. In one embodiment, NCG-associated data vectors, CGN-associated data vectors, and combined data vectors (i.e., data vectors containing terminal motifs of CGs) are used, which can be obtained from HCC-specific hypermethylated and hypomethylated CpGs of 55 non-HCC cases and 34 HCC cases, respectively.

[0539] In the NCG-related data vector, only the NCG motif frequency is used. For example, the frequency of ACG is calculated as the amount of ACG motif / the total amount of NCG motif. The sum of the frequencies of ACG, TCG, CCG, and GCG is 100%. Similar calculations are applied to the CGN-related data vector. For example, the frequency of CGA is calculated as the number of CGA motifs / the total amount of CGN motifs. The sum of the frequencies of CGA, CGT, CGC, and CGG is 100%. The data vector of the combination of all CGN and NCG motifs is calculated. For example, the frequency of CGA is calculated as the number of CGA motifs / the total amount of CGN and NCG motifs. The sum of the frequencies of ACG, TCG, CCG, GCG, CGA, CGT, CGC, and CGG is 100%. These data sets are used to construct support vector machine (SVM) models respectively.

[0540] Figures 75A to 75BDemonstrate the performance of HCC detection based on CG-containing 3-mer terminal motifs from HCC-specific hypermethylated and hypomethylated CpGs. Figure 75A As shown in Figure 3, the SVM models trained based on the NCG-related data vectors, CGN-related data vectors, and combined data vectors from HCC-specific hypermethylated CpGs achieved AUCs of 0.85, 0.89, and 0.98, respectively, to distinguish HCC patients from non-HCC cases. Figure 75B As shown, for HCC-specific hypomethylated CpGs, those models achieved AUCs of 0.94, 0.90, and 0.98, respectively. Figure 74B The accuracy performance is better.

[0541] Some embodiments may combine feature vectors from hypermethylated and hypomethylated sites for training a machine learning model, such as an SVM.

[0542] Figure 76 ROC analysis is presented for distinguishing patients with HCC from those without HCC using a combined data vector of HCC-specific hypermethylated and hypomethylated CpGs. To further maximize the utility of the terminal motifs from HCC-specific methylated CpGs, a combined data vector of HCC-specific hypermethylated and hypomethylated CpGs was integrated and used to train an SVM model, which achieved an AUC of 0.99 for distinguishing HCC patients from non-HCC cases. 3. Use cutting mode

[0543] We also investigated whether cleavage patterns from tissue-specific hypermethylated and hypomethylated CpG sites could be used to detect subjects with pathological or physiological conditions. As explained above, the cleavage pattern includes two or more positions including a terminal motif of a CGN, for example, one at position 0. a) Example Workflow

[0544] For illustration purposes, pregnancy is used as a model. Pregnancy can be considered as a pathological or physiological state or similar to a pathological or physiological state because another tissue type with a characteristic different from that of the host is growing. For example, fetuses and tumors are all generally hypomethylated, and some sites are hypermethylated, such as the sites of tumor suppressor genes. Therefore, if an object can be correctly classified as pregnant with a fetus, wherein such tissue has a specific methylation pattern, then the embodiment can be used to detect lesions related to other tissue types using hypomethylated or hypermethylated CpG sites. Other analyses can be a group of CpG sites (e.g., CpG islands). A region can be such one or more CpG sites or include such one or more CpG sites.

[0545] Figure 77Example marker selection and data collection according to embodiments of the present disclosure are shown. Placenta-specific hypermethylated sites defined by those CpG sites with a methylation density greater than 70% in placental tissue but less than 30% in buffy coat samples are identified. In other embodiments, these methylation density cutoffs can be changed to achieve different diagnostic sensitivities and / or specificities, as described in various embodiments of the present disclosure. Hypomethylation can be defined as, but not limited to, less than 10%, less than 20%, less than 30%, less than 40%, less than 50%, etc. Hypermethylation can be defined as, but not limited to, more than 40%, more than 50%, more than 60%, more than 70%, more than 80%, more than 90%, etc. The cutoff value can be a range, such as, but not limited to, 20% to 30%, 30% to 40%, 40% to 50%, 60% to 70%, 70% to 80%, etc. Placenta-specific hypomethylated sites were defined by those CpG sites with a methylation density less than 30% in placental tissue but more than 70% in buffy coat samples.

[0546] For data collection, cleavage patterns of placenta-specific hypermethylated and hypomethylated CpG sites were determined. Cleavage patterns were determined using the techniques described herein. Such cleavage patterns were determined for 30 pregnant samples (median paired-end sequencing reads: 206 million (IQR: 142 million to 232 million)) and 9 non-pregnant healthy female cases (median paired-end sequencing reads: 397 million (IQR: 350 million to 424 million)). In this example embodiment, a cleavage pattern can be generated by merging data for all sites of a particular type, for example, merging data for all hypomethylated sites for one cleavage pattern and data for all hypermethylated sites for another cleavage pattern.

[0547] Figure 78 This article demonstrates a workflow for generating cleavage patterns for various subjects for training a machine learning model, according to an embodiment of the present disclosure. In an example using hypermethylated and hypomethylated sites presented in a measurement window, a total of 24 features are present. The machine learning model can then distinguish between different subjects based on these features. Results are provided below. b) Clustering results

[0548] In the results below, clustering was performed using principal component analysis (PCA).Other embodiments may use other types of machine learning, supervised or unsupervised, or a combination thereof, as described elsewhere in this disclosure.

[0549] Figure 79A 、 79B and Figure 80According to an embodiment of the present disclosure, by using a methylation site from placenta specific hypermethylation ( Figure 79A ), placenta-specific hypomethylated sites ( Figure 79B ) or a combination of placenta-specific hypermethylated and hypomethylated sites ( Figure 80 ) cutting pattern on pregnant and non-pregnant subjects. PCA analysis identifies the linear combination of features that provides the best separation between the two types of subjects. Support vector machines can provide similar results. PCA allows observations along the two identified principal dimensions.

[0550] like Figure 79A As shown in , the principal components (PCs) based on the cleavage patterns from placenta-specific hypermethylated sites (i.e., PC1 and PC2) can classify pregnant and non-pregnant subjects into two clusters, where only two non-pregnant subjects overlap with the pregnant subjects. The principal components based on the cleavage patterns from placenta-specific hypomethylated sites (i.e., PC1 and PC2) showed better discrimination between pregnant and non-pregnant cases without any overlap between the two groups ( Figure 79B ).

[0551] Furthermore, PCs based on cleavage patterns from placenta-specific hypermethylated and hypomethylated sites (i.e., PC1 and PC2) showed complete separation between pregnant and non-pregnant cases ( Figure 80 ). The data suggest that selective analysis of cleavage patterns originating from hypomethylated and / or hypermethylated CpG sites will improve classification power for categorizing different pathological and physiological states. c) Example use cases

[0552] Cleavage patterns derived from tissue-specific hypomethylated or hypermethylated CpG sites will be applicable to the detection of various cancers or other diseases. For example, cleavage patterns derived from colon-specific, breast-specific, or prostate-specific hypomethylated or hypermethylated CpG sites can be used for colon cancer, breast cancer, or prostate cancer diagnosis, respectively. Therefore, cleavage patterns will be applicable to the prediction and / or monitoring of cancer patients.

[0553] For example, cleavage patterns derived from tissue-specific CpGs can be used to infer the DNA contribution of different tissues to the plasma cfDNA pool, such as the pro...

Claims

1. A method for measuring methylation of a CpG site in a subject's genome using free DNA molecules, the method comprising: Analyzing a plurality of cell-free DNA molecules from a biological sample of the subject, wherein analyzing the cell-free DNA molecules comprises: determining a genomic position in a reference genome corresponding to at least one end of the free DNA molecule; determining a first amount of free DNA molecules having an end located at a first position within a window around the CpG site, the first position being between positions -1 and +1 of the window; and A classification of methylation of the CpG site in the genome of the subject is determined using the first quantity.

2. The method of claim 1, further comprising: A second amount of free DNA molecules having an end located at a second position within the window around the CpG site is determined, the second position being different from the first position, wherein the classification is determined using the first amount and the second amount.

3. The method of claim 2, wherein the first position is position 0, and wherein the second position is at +1 or -1 from the CpG site.

4. The method of claim 2, wherein the window is at least -2 to +2 from the CpG site.

5. The method of claim 2, wherein determining the classification comprises: determining a separation value using the first amount and the second amount; and The separation values ​​are compared to calibration values ​​determined using free DNA molecules from one or more calibration samples and located at CpG sites with known classification.

6. The method of claim 5, wherein the classification is a quantitative value, and wherein comparing the separation value to the calibration value comprises comparing the separation value to a calibration function.

7. The method of claim 6, wherein the quantitative value is in the range of 30% or less.

8. The method of claim 2, wherein the CpG site is a first CpG site and the classification is a first classification, and wherein the window includes a second CpG site and at least two positions other than the first CpG site and the second CpG site, and the method further comprises: determining a third amount of free DNA molecules having ends located at the second CpG site; For each of the at least two positions within the window: determining respective amounts of free DNA molecules having ends at said positions, thereby determining respective amounts including said second amount, wherein said at least two positions include said first position; generating a feature vector including the respective quantities, the first quantity and the third quantity; and The feature vector is input into a machine learning model as part of determining the classification of the first CpG site and determining a second classification of the second CpG site, wherein the machine learning model is trained using free DNA molecules located within a window around CpG sites with known classification.

9. The method of claim 1, wherein the window comprises at least two positions other than the first position, the method further comprising: For each of the at least two positions within the window: determining respective amounts of free DNA molecules having ends located at said positions, wherein said at least two positions include said first position; and The first amount of free DNA molecules ending at the CpG site is compared to the respective amounts of free DNA molecules ending at the position as part of determining the classification.

10. The method of claim 1, wherein the window comprises at least two positions other than the first position, the method further comprising: For each of the at least two positions within the window: determining the respective amount of free DNA molecules with their ends located at said positions, thereby determining a plurality of respective amounts; generating a feature vector including the respective quantities and the first quantity; and The feature vector is input into a machine learning model as part of determining the classification, wherein the machine learning model is trained using free DNA molecules located within a window around CpG sites with known classification.

11. The method of claim 10, wherein the feature vectors form a matrix with each row corresponding to a base of a strand, and wherein the columns include non-zero quantities in the rows corresponding to the base at each position.

12. The method of claim 10, wherein the machine learning model is a convolutional neural network.

13. The method of claim 10, wherein the first position and the at least two positions include all positions within the window, the window being at least +4 to -4 from the CpG site.

14. The method of claim 10, wherein the machine learning model uses sequence context within the window.

15. The method of claim 14, wherein the machine learning model is trained on the sequence context within the window.

16. The method of claim 14, wherein the feature vector comprises the sequence context.

17. A method as claimed in claim 1, wherein the classification indication of methylation is for the free DNA molecule at the CpG site, and the CpG site is in a high methylation state or a low methylation state, wherein the high methylation state indicates that the methylation density exceeds a first critical value, and the first critical value is at least 70%, and wherein the low methylation state indicates that the methylation density is lower than a second critical value, and the second critical value is 30% or less.

18. The method of claim 1, wherein the first amount is normalized.

19. The method of claim 18, wherein the normalization uses the number of free DNA molecules whose ends are located within the region including the CpG site.

20. The method of claim 18, wherein the normalization uses the number of free DNA molecules covering the CpG sites.

21. The method of claim 18, wherein the normalization uses the average or median depth of free DNA molecules in the region including the CpG site.

22. The method of claim 1 or claim 2, further comprising: Another classification of methylation of a different CpG site in the genome of the subject is determined using the first amount and optionally the second amount, the different CpG site being within 600 nt downstream of the 5' end of the CpG site.

23. A method for measuring methylation of a region in a subject's genome, the method comprising: Analyzing a plurality of cell-free DNA molecules from a biological sample of the subject, wherein analyzing the cell-free DNA molecules comprises: determining a terminal sequence motif of at least one end of the free DNA molecule, wherein the end of the free DNA molecule has a first position located at an outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position, wherein the plurality of free DNA molecules are located in the region of the reference genome; A first quantity of the first set of one or more terminal sequence motifs is determined, wherein: The one or more terminal sequence motifs of the first group have a C at the first position and a G at the second position, or The one or more terminal sequence motifs of the first group have a C at the second position and a G at the third position; and The methylation classification of the region in the genome of the subject is determined by comparing the first amount to a calibration value determined using free DNA molecules from one or more calibration samples and located at CpG sites with known classification.

24. The method of claim 23, wherein the region is a portion of the genome of the subject, and wherein the plurality of cell-free DNA molecules are from one or more specific regions of the reference genome.

25. The method of claim 24, wherein the region comprises a sequence variant or copy number aberration determined from the plurality of free DNA molecules, the method further comprising: Based on the methylation classification of the region, it is determined whether the sequence variant or the copy number aberration is from a tumor.

26. The method of claim 24, wherein at least one of the one or more specific regions corresponds to a specific genotype or haplotype comprising one or more alleles.

27. The method of claim 26, wherein the methylation classification of the region is a first methylation level corresponding to a first haplotype of a female pregnant with a fetus, the method further comprising: determining a second methylation level corresponding to the second haplotype of the female using a second amount of the first set of one or more terminal sequence motifs from free DNA molecules of a second haplotype; and A genetic haplotype of the fetus is determined based on the first methylation level and the second methylation level.

28. The method of claim 26, further comprising: The grade of the lesion of the subject in the one or more specific regions is determined based on the classification.

29. The method of claim 28, wherein the grade of the lesion comprises the abnormal degree of a physiological process or a copy number aberration.

30. The method of claim 23, wherein the first set of one or more terminal sequence motifs has a C at the first position and a G at the second position, the method further comprising: determining a second amount of one or more terminal sequence motifs of a second group having a C at the second position and a G at the third position, Wherein determining the classification comprises normalizing the first quantity with the second quantity to obtain a normalized first quantity that is compared with the calibration value.

31. The method of claim 30, wherein the first group of one or more terminal sequence motifs comprises CGA, CGC, CGG and CGT, and wherein the second group of one or more terminal sequence motifs comprises ACG, CCG, GCG and TCG.

32. The method of claim 23, further comprising: determining an expected amount of one or more terminal sequence motifs of the first group based on a reference sequence of the reference genome in the region, Wherein determining the classification comprises normalizing the first quantity with the expected quantity to obtain a normalized first quantity that is compared to the calibration value.

33. The method of claim 23, further comprising: determining the total amount of sequence motifs of the plurality of free DNA molecules, Wherein determining the classification comprises normalizing the first amount with the total amount to obtain a normalized first amount that is compared with the calibration value.

34. The method of claim 23, wherein the region is the reference genome.

35. The method of claim 23, wherein the region is 10 Mb or less of a specific chromosome.

36. The method of claim 23, wherein the first set of one or more terminal sequence motifs has a C at the first position and a G at the second position, the method further comprising: determining an individual amount of each 3-mer terminal sequence motif having a C at the first position and a G at the second position, thereby determining a plurality of individual amounts; generating a feature vector including the quantities, wherein the quantities include the first quantity; and The feature vector is input into a machine learning model as part of determining the methylation classification of the region in the genome of the subject, wherein the machine learning model is trained using free DNA molecules from the one or more calibration samples.

37. The method of claim 36, wherein the feature vector comprises individual quantities of all 3-mer terminal sequence motifs.

38. The method of claim 23, wherein analyzing free DNA molecules comprises: The position of the free DNA molecule in the reference genome is determined.

39. A method for measuring methylation of CpG sites of a test free DNA molecule from a biological sample of a subject, the method comprising: receiving one or more sequence reads of the test free DNA molecule, wherein the test free DNA molecule comprises one or more CpG sites; Determine a sequence context of the test free DNA molecule using the one or more sequence reads, wherein the sequence context includes sequence information for each nucleotide from the test free DNA molecule; determining a first end sequence motif of a first end of the test free DNA molecule using the one or more sequence reads; Loading a machine learning model into the memory of a computer system, wherein the machine learning model is trained using a training set of training free DNA molecules having CpG sites with known methylation status, wherein the training set includes sequence background and terminal sequence motifs of the training free DNA molecules; and The methylation level of the one or more CpG sites of the test free DNA molecule is determined using the machine learning model, wherein (1) the sequence background of the test free DNA molecule and (2) the first terminal sequence motif are input into the machine learning model.

40. The method of claim 39, further comprising: A second end sequence motif of a second end of the test free DNA molecule is determined using the one or more sequence reads, wherein the second end sequence motif is further input into the machine learning model.

41. The method of claim 39, wherein the first terminal sequence motif is one-hot encoded prior to input into the machine learning model.

42. The method of claim 39, wherein the sequence background comprises the entire sequence of the test cell-free DNA molecule.

43. A method as claimed in claim 39, wherein the sequence background includes a K-mer matrix specifying multiple examples of each K-mer in the test free DNA molecule, where K is an integer.

44. The method of claim 43, wherein K is equal to 2, and wherein the K-mer matrix is ​​a dinucleotide matrix specifying multiple examples of dinucleotide pairs in the test free DNA molecules.

45. The method of claim 39, wherein the machine learning model comprises a convolutional neural network.

46. ​​The method of claim 39, wherein the test free DNA molecule comprises a plurality of CpG sites, and wherein the methylation level is a percentage of the plurality of CpG sites that are methylated.

47. The method of claim 39, further comprising: determining the size of the test free DNA molecule; and The size of the test free DNA molecule is input into the machine learning model as part of determining the methylation level of the one or more CpG sites of the test free DNA molecule.

48. The method of claim 47, wherein the one or more sequence reads are one sequence read spanning the test free DNA molecule, and wherein determining the size of the test free DNA molecule comprises counting the number of bases in the one sequence read.

49. A method as claimed in claim 47, wherein the one or more sequence reads include two sequence reads corresponding to the two ends of the test free DNA molecule, and wherein determining the size of the test free DNA molecule includes aligning the two sequence reads with a reference genome.

50. A method of analyzing a biological sample to determine the grade of a lesion in the biological sample of a subject, the biological sample comprising cell-free DNA, the method comprising: analyzing a plurality of free DNA molecules from the biological sample to obtain sequence reads, wherein the sequence reads include end sequences corresponding to ends of the plurality of free DNA molecules; For each of the plurality of free DNA molecules, determining a sequence motif for each of one or more ends of the free DNA molecule, wherein the ends of the free DNA molecules have a first position located at an outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position; determining a first quantity of one or more terminal sequence motifs of a first group of the plurality of free DNA molecules, wherein: The one or more terminal sequence motifs of the first group have a C at the first position and a G at the second position, or The one or more terminal sequence motifs of the first group have a C at the second position and a G at the third position; and A classification of the lesion grade of the subject is determined based on a comparison of the first quantity with a reference value.

51. The method of claim 50, wherein analyzing the plurality of free DNA molecules comprises measuring a size of each of the plurality of free DNA molecules, and wherein the first amount is determined for a first group of free DNA molecules within a first size range.

52. The method of claim 51, wherein the first size range corresponds to sizes smaller than a size cutoff value.

53. The method of claim 51, wherein the first size range corresponds to sizes greater than a first size cutoff.

54. The method of claim 53, further comprising: determining a second amount of one or more terminal sequence motifs of a second group of the plurality of free DNA molecules, wherein: The one or more terminal sequence motifs of the second group have a C at the first position and a G at the second position, or the one or more terminal sequence motifs of the second group have a C at the second position and a G at the third position, wherein the second amount is determined for a second group of free DNA molecules within a second size range, the second size range corresponding to sizes less than a second size cutoff value, and wherein the classification is determined using the first amount and the second amount.

55. The method of claim 54, wherein the first amount is determined using one or more terminal sequence motifs of the first group having a C at the first position and a G at the second position, and wherein the second amount is determined using one or more terminal sequence motifs of the second group having a C at the first position and a G at the second position, the method further comprising: determining a third amount of said first group of free DNA molecules having a third group of terminal sequence motifs having a C at said second position and a G at said third position, and A fourth amount of the second group of free DNA molecules having a fourth group of terminal sequence motifs having a C at the second position and a G at the third position is determined, wherein the classification is further determined using the third amount and the fourth amount.

56. A method as claimed in claim 54, wherein the classification uses a ratio of the first quantity to the second quantity, uses a difference between the first quantity and the second quantity, or uses a machine learning model that receives the first quantity and the second quantity as separate inputs.

57. The method of claim 50, wherein the first amount is determined for a first group of free DNA molecules each located within one or more regions each enriched in 5hmC or each depleted in 5hmC of a specific tissue type, and wherein the lesion belongs to the specific tissue type.

58. The method of claim 57, wherein the specific tissue type is HCC cancer tissue.

59. The method of claim 57, wherein the first group of free DNA molecules are each located in one or more regions each enriched in 5hmC, the method further comprising: Determine a second quantity of the one or more terminal sequence motifs of the first group of the plurality of free DNA molecules, each located within one or more regions of the specific tissue type that are each lacking 5hmC, wherein the classification is determined using the first quantity and the second quantity.

60. The method of claim 50, wherein the first amount is determined for a first group of free DNA molecules each located within one or more respective hypermethylated or hypomethylated regions of a specific tissue type, and wherein the lesion belongs to the specific tissue type.

61. The method of claim 50, wherein the plurality of free DNA molecules are from a first region of a plurality of regions of a reference genome, and wherein the comparison of the first amount with the reference value indicates whether the first amount of the first region is abnormal, the method further comprising: determining a number of the plurality of regions that are abnormal; and The number of abnormal regions is compared to a cutoff value to determine the classification of the lesion grade.

62. The method of claim 61, wherein: For each of the plurality of regions, the respective amount of the one or more terminal sequence motifs of the first set is compared to a region-specific reference value.

63. The method of claim 50, wherein the plurality of free DNA molecules are from one or more specific regions of a reference genome, and wherein the lesions are in the one or more specific regions.

64. The method of claim 63, wherein at least one of the one or more specific regions corresponds to a specific genotype or haplotype comprising one or more alleles.

65. The method of claim 63, wherein the grade of the lesion comprises copy number aberrations.

66. The method of claim 50, wherein the first set of one or more terminal sequence motifs comprises CGN, wherein N comprises any nucleotide at the third position.

67. The method of claim 50, wherein the first set of one or more terminal sequence motifs has a C at the first position and a G at the second position, the method further comprising: determining a second amount of a second group of terminal sequence motifs having a C at the second position and a G at the third position, Wherein determining the classification comprises normalizing the first quantity with the second quantity to obtain a normalized first quantity for comparison with the reference value.

68. The method of claim 67, wherein the second set of terminal sequence motifs comprises NCG, wherein N comprises any nucleotide at the first position.

69. The method of claim 50, further comprising: determining the total amount of sequence motifs of the plurality of free DNA molecules, Wherein determining the classification comprises normalizing the first amount with the total amount to obtain a normalized first amount for comparison with the reference value.

70. The method of claim 50, wherein the lesion is cancer.

71. The method of claim 50, wherein the disorder is an abnormal degree of a physiological process, wherein the physiological process is pregnancy, aging, development, or growth.

72. The method of claim 50, wherein the first set of one or more terminal sequence motifs has a C at the first position and a G at the second position, the method further comprising: determining an individual amount of each 3-mer terminal sequence motif having a C at the first position and a G at the second position, thereby determining a plurality of individual amounts; generating a feature vector including the quantities, wherein the quantities include the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the lesion grade of the subject, wherein the machine learning model is trained using free DNA molecules in training samples with known classification.

73. The method of claim 72, further comprising: An individual additional quantity for each 3-mer terminal sequence motif having a C at the second position and a G at the third position is determined, thereby determining a plurality of individual additional quantities, wherein the feature vector includes the individual additional quantities.

74. The method of claim 72, wherein the feature vector comprises individual quantities of all 3-mer terminal sequence motifs.

75. The method of claim 50, wherein the first set of one or more terminal sequence motifs has a C at the second position and a G at the third position, the method further comprising: determining an individual amount of each 3-mer terminal sequence motif having a C at the second position and a G at the third position, thereby determining a plurality of individual amounts; generating a feature vector including the quantities, wherein the quantities include the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the lesion grade of the subject, wherein the machine learning model is trained using free DNA molecules in training samples with known classification.

76. A method of analyzing a biological sample to determine the grade of a lesion in the biological sample of a subject, the biological sample comprising cell-free DNA, the method comprising: (a) analyzing a plurality of free DNA molecules from the biological sample of the subject, wherein analyzing the free DNA molecules comprises: Determining the locations of the cell-free DNA molecules in a reference genome, wherein the plurality of cell-free DNA molecules are located in a first group of one or more regions of the reference genome, and wherein each of the one or more regions of the first group is hypomethylated or each is hypermethylated in a first tissue type; and Determining a terminal sequence motif of at least one end of the free DNA molecule, wherein the end of the free DNA molecule has a first position located at the outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position; (b) determining a first quantity of the first set of one or more terminal sequence motifs, wherein: The one or more terminal sequence motifs of the first group have a C at the first position and a G at the second position, or The one or more terminal sequence motifs of the first group have a C at the second position and a G at the third position; and (c) determining a classification of the grade of the lesion in the first tissue type of the subject based on a comparison of the first quantity with a reference value.

77. The method of claim 76, wherein the first tissue type is a cancer tissue type.

78. The method of claim 77, wherein the biological sample is urine, and wherein the cancer tissue type is selected from the group consisting of bladder cancer, kidney cancer, and prostate cancer.

79. The method of claim 77, wherein the biological sample is plasma or serum, and wherein the cancer tissue type is selected from the group consisting of liver cancer, colon cancer, lung cancer, and breast cancer.

80. The method of claim 76, wherein the classification uses differentially methylated regions of a plurality of tissue types including the first tissue type, the method further comprising: Analyze other groups of free DNA molecules from the biological sample, wherein each group of free DNA molecules is located in one or more regions of each group of the reference genome, and each of the one or more regions of each group in each tissue type of the multiple tissue types is hypomethylated or hypermethylated.

81. A method as claimed in claim 80, wherein a particular set of one or more regions includes multiple regions of multiple region types, wherein a first region type is differentially methylated relative to other tissues of the multiple tissue types, and wherein a second region type is differentially methylated relative to at least one other tissue of the multiple tissue types but not all other tissues.

82. The method of claim 81, wherein another specific set of one or more regions in the cancerous tissue type of the first tissue and the healthy first tissue are all hypomethylated or all hypermethylated.

83. A method as claimed in claim 80, wherein the classification of the lesion grade uses a first machine learning model, and wherein the amount of one or more terminal sequence motifs in the first group is used to train one or more other machine learning models using training samples having the lesion in one or more other tissue types of the multiple tissue types.

84. The method of claim 80, wherein the classification of the lesion grade uses a multi-class machine learning model that provides a probability that each of the plurality of tissue types has the lesion.

85. The method of claim 80, wherein the plurality of tissue types comprises cancerous tissue types and non-cancerous tissue types.

86. The method of claim 76, further comprising: identifying a second set of CpG sites in the reference genome that are either all hypomethylated or all hypermethylated across multiple tissue types; and determining a second amount of free DNA molecules whose ends are located around any one of the CpG sites in the second group, Wherein determining the classification of the lesion grade comprises normalizing the first quantity with the second quantity to obtain a normalized first quantity for comparison with the reference value.

87. The method of claim 76, wherein the first set of one or more terminal sequence motifs define the first set of one or more CpG sites within the first set of one or more regions, the method further comprising: For each of at least two positions within a window around any one of the one or more CpG sites in the first group: determining the respective amount of free DNA molecules with their ends located at said positions, thereby determining a plurality of respective amounts; generating a feature vector including the respective quantities and the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the lesion grade in the first tissue type of the subject, wherein the machine learning model is trained using free DNA molecules from training samples with known classification.

88. The method of claim 87, wherein for the first tissue type, the first set of one or more CpG sites are all hypomethylated, and wherein the feature vector is a first feature vector, the method further comprising: generating a second feature vector using the one or more CpG sites of the second group that are all hypermethylated; and The second feature vector and the first feature vector are input into the machine learning model as part of determining the classification of the grade of the lesion in the first tissue type of the subject.

89. The method of claim 76, wherein the first set of one or more terminal sequence motifs define a first set of CpG sites within the first set of one or more regions, and wherein determining the first amount comprises: Respective amounts of free DNA molecules having a terminal end at each of the CpG sites in the first group are determined, wherein determining the classification of the grade of the lesion comprises comparing each of the respective amounts with respective reference values.

90. The method of claim 89, wherein each of the individual quantities is compared to the individual reference values ​​by inputting the individual quantities into a machine learning model as part of a feature vector.

91. The method of claim 90, further comprising, for each CpG site of the first set of one or more CpG sites: For each of at least two positions within a window around the CpG site: determining individual amounts of free DNA molecules having ends located at said positions, thereby determining a plurality of individual amounts; and The feature vector includes the respective quantities of the at least two positions and the respective quantities of the CpG sites.

92. The method of claim 76, wherein the first group of one or more regions is a plurality of regions.

93. The method of claim 76, wherein the first set of one or more terminal sequence motifs comprises CGN, wherein N comprises any nucleotide at the third position.

94. The method of claim 76, further comprising: determining a second amount of a second group of terminal sequence motifs having a C at the second position and a G at the third position, Wherein determining the classification comprises normalizing the first quantity with the second quantity to obtain a normalized first quantity for comparison with the reference value.

95. The method of claim 94, wherein the second set of terminal sequence motifs comprises NCG, wherein N comprises any nucleotide at the first position.

96. The method of claim 76, further comprising: determining the total amount of sequence motifs of the plurality of free DNA molecules, Wherein determining the classification comprises normalizing the first amount with the total amount to obtain a normalized first amount for comparison with the reference value.

97. The method of claim 76, wherein the lesion is cancer.

98. The method of claim 76, wherein the disorder is an abnormal degree of a physiological process, wherein the physiological process is pregnancy, aging, development, or growth.

99. The method of claim 76, further comprising: (a), (b) and (c) are repeated for one or more regions of an additional set or sets of the reference genome, and wherein the additional set or sets are all hypomethylated or all hypermethylated in one or more other tissue types, thereby determining the cancer type of the subject.

100. The method of claim 76, wherein the first set of one or more terminal sequence motifs includes all sequence motifs having a C at the first position and a G at the second position, the method further comprising: determining an individual amount of each 3-mer terminal sequence motif having a C at the first position and a G at the second position, thereby determining a plurality of individual amounts; generating a feature vector including the quantities, wherein the quantities include the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the lesion grade in the first tissue type of the subject, wherein the machine learning model is trained using free DNA molecules in training samples with known classification.

101. The method of claim 100, further comprising: An individual additional quantity for each 3-mer terminal sequence motif having a C at the second position and a G at the third position is determined, thereby determining a plurality of individual additional quantities, wherein the feature vector includes the individual additional quantities.

102. The method of claim 100, wherein the feature vector comprises individual quantities of all 3-mer terminal sequence motifs.

103. The method of claim 76, wherein the first set of one or more terminal sequence motifs includes all sequence motifs having a C at the second position and a G at the third position, the method further comprising: determining an individual amount of each 3-mer terminal sequence motif having a C at the second position and a G at the third position, thereby determining a plurality of individual amounts; generating a feature vector including the quantities, wherein the quantities include the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the lesion grade of the subject, wherein the machine learning model is trained using free DNA molecules in training samples with known classification.

104. The method of claim 100 or claim 103, wherein the classification uses differentially methylated regions of a plurality of tissue types including the first tissue type, the method further comprising: Analyzing other groups of free DNA molecules from the biological sample, wherein each group of free DNA molecules is located in one or more regions of each group of the reference genome, and each of the one or more regions of each group in each tissue type of the multiple tissue types is hypomethylated or hypermethylated, wherein the feature vector forms a matrix in which each row corresponds to one of the one or more regions of the respective groups of the reference genome, and wherein the columns correspond to the respective quantities of each 3-mer terminal sequence motif of the first group, and wherein the machine learning model is a convolutional neural network.

105. A method of analyzing a biological sample comprising a mixture of free DNA molecules, said mixture comprising free DNA molecules from said subject and from a virus, to determine the grade of a lesion in a subject from which said biological sample was obtained, said method comprising: (a) analyzing a plurality of viral free DNA molecules from the biological sample of the subject, wherein analyzing each of the set of viral free DNA molecules of the virus comprises: Determining the locations of the viral free DNA molecules in a reference genome of the virus, wherein the group of viral free DNA molecules is located at one or more CpG sites of a first group of the reference genome; and Determining a terminal sequence motif of at least one end of the viral free DNA molecule, thereby determining a plurality of terminal sequence motifs, wherein the end of the viral free DNA molecule has a first position located at the outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position; (b) determining a first amount of said set of viral free DNA molecules of said virus having a terminus located at any one or more of positions -1 to +1 relative to any one of said first set of one or more CpG sites using said terminal sequence motif; and (c) determining the classification of the lesion grade of the subject based on a comparison of the first quantity with a reference value.

106. The method of claim 105, wherein each of the first set of one or more CpG sites has a methylation level within a specified range in a subject having the lesion.

107. A method as claimed in claim 105, wherein a first training set of samples from subjects having the lesion and a second training set of samples from subjects not having the lesion are used to identify each of the one or more CpG sites in the first group, wherein the first quantity at each site in the one or more CpG sites in the first group differs by at least a specified percentage between the first training set and the second training set.

108. The method of claim 105, wherein each of the first set of one or more CpG sites is differentially methylated in subjects having the lesion relative to subjects not having the lesion.

109. The method of claim 105, wherein the first amount is the amount of a first group of one or more terminal sequence motifs having a C at the first position and a G at the second position.

110. The method of claim 109, wherein the first set of one or more terminal sequence motifs comprises CGN, wherein N comprises any nucleotide at the third position.

111. The method of claim 109, further comprising: determining a second amount of a second group of terminal sequence motifs having a C at the second position and a G at the third position, Wherein determining the classification comprises normalizing the first quantity with the second quantity to obtain a normalized first quantity for comparison with the reference value.

112. The method of claim 111, wherein the second set of terminal sequence motifs comprises NCG, wherein N comprises any nucleotide at the first position.

113. The method of claim 105, wherein the first amount is the amount of one or more terminal sequence motifs of the first group having a C at the second position and a G at the third position.

114. The method of claim 105, wherein the first amount is the sum of the viral free DNA molecules whose ends are located at the -1 or +1 position of the one or more CpG sites of the first group.

115. The method of claim 114, wherein the first group of one or more CpG sites are each hypomethylated.

116. The method of claim 105, further comprising: For each of at least two positions within a window around any one of the one or more CpG sites in the first group: determining the respective amount of free DNA molecules with their ends located at said positions, thereby determining a plurality of respective amounts; generating a feature vector including the respective quantities and the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the lesion grade, wherein the machine learning model is trained using free DNA molecules from training samples with known classification.

117. The method of claim 116, wherein each of the first set of one or more CpG sites is hypomethylated in a subject having the lesion.

118. The method of claim 117, wherein the feature vector is a first feature vector, the method further comprising: generating a second feature vector using the one or more CpG sites of the second group that are all hypermethylated; and The second feature vector and the first feature vector are input into the machine learning model as part of the classification to determine the lesion grade.

119. The method of claim 105, wherein the first set of one or more CpG sites is a first set of CpG sites, wherein determining the first amount comprises: determining the respective amounts of free DNA molecules having a terminus at each of the CpG sites in the first set, and The classification wherein determining the grade of the lesion comprises comparing each of the respective quantities with respective reference values.

120. The method of claim 119, wherein each of the individual quantities is compared to the individual reference values ​​by inputting the individual quantities into a machine learning model as part of a feature vector.

121. The method of claim 120, further comprising, for each CpG site of the first set of one or more CpG sites: For each of at least two positions within a window around the CpG site: determining individual amounts of free DNA molecules having ends located at said positions, thereby determining a plurality of individual amounts; and The feature vector includes the respective quantities of the at least two positions and the respective quantities of the CpG sites.

122. The method of claim 105, further comprising: determining the total amount of sequence motifs of the plurality of viral free DNA molecules, Wherein determining the classification comprises normalizing the first amount with the total amount to obtain a normalized first amount for comparison with the reference value.

123. The method of claim 105, wherein the lesion is cancer.

124. The method of claim 105, further comprising: determining an individual amount of each 3-mer terminal sequence motif having a C at the first position and a G at the second position, thereby determining a plurality of individual amounts; generating a feature vector including the quantities, wherein the quantities include the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the lesion grade, wherein the machine learning model is trained using free DNA molecules in training samples with known classification.

125. The method of claim 124, further comprising: An individual additional quantity for each 3-mer terminal sequence motif having a C at the second position and a G at the third position is determined, thereby determining a plurality of individual additional quantities, wherein the feature vector includes the individual additional quantities.

126. The method of claim 124, wherein the feature vector comprises individual quantities of all 3-mer terminal sequence motifs.

127. The method of claim 105, wherein analyzing the plurality of viral free DNA molecules comprises determining a size of each of the plurality of viral free DNA molecules, the method further comprising: determining a statistical value of the size distribution of the plurality of viral free DNA molecules, and Wherein determining the classification of the lesion grade of the object further comprises comparing the statistical value with a size reference value.

128. The method of claim 105 or claim 127, further comprising: determining the amount of the plurality of viral free DNA molecules, and Wherein determining the classification of the lesion grade of the subject further comprises comparing the amount to a count reference value.

129. A method for measuring the proportional concentration of DNA from a first tissue type in a biological sample of a subject, the biological sample comprising cell-free DNA, the method comprising: Analyzing a plurality of cell-free DNA molecules from the biological sample of the subject, wherein analyzing the cell-free DNA molecules comprises: determining a genomic position in a reference genome corresponding to at least one end of the free DNA molecule; identifying a first set of CpG sites in the reference genome that all have a first differential methylation for the first tissue type, wherein the first set of CpG sites includes one or more CpG sites; determining a first amount of free DNA molecules having an end located at one of the 0 or -1 positions of any one of the first set of CpG sites; and The proportional concentration of DNA from the first tissue type in the biological sample is determined by comparing the first amount to a calibration value determined from one or more calibration samples having a known proportional concentration of DNA from the first tissue type.

130. The method of claim 129, wherein the first differential methylation is that all CpG sites of the first group are hypomethylated, or wherein the first differential methylation is that all CpG sites of the first group are hypermethylated.

131. The method of claim 130, further comprising: identifying a second set of CpG sites in the reference genome that are either all hypomethylated or all hypermethylated across multiple tissue types; and determining a second amount of free DNA molecules whose ends are within a window around any one of the CpG sites in the second group, Wherein determining the proportional concentration comprises normalizing the first amount with the second amount to obtain a normalized first amount that is compared to the calibration value.

132. The method of claim 130, further comprising: identifying a second set of CpG sites in the reference genome that are all hypomethylated or all hypermethylated for a second tissue type, wherein the second set of CpG sites comprises one or more CpG sites; determining a second amount of free DNA molecules whose ends are located at any one of the CpG sites in the second group; and A change in the proportional concentration of DNA from the second tissue type in the biological sample is determined by comparing the second amount to another calibration value determined from one or more other calibration samples.

133. The method of claim 129, wherein the first differential methylation is that all CpG sites of the first group are enriched with 5hmC, or wherein the first differential methylation is that all CpG sites of the first group are depleted of 5hmC.

134. The method of claim 129, wherein determining the first amount comprises: determining the respective amounts of free DNA molecules having their ends at position 0 of each of the CpG sites of the first set, and Wherein determining the proportional concentration of DNA from the first tissue type comprises comparing each of the respective amounts to a respective calibration value.

135. A method as claimed in claim 134, wherein each of the individual quantities is compared with the individual calibration values ​​by inputting the individual quantities into a machine learning model as part of a feature vector.

136. The method of claim 135, further comprising, for each of the one or more CpG sites: For each of at least two positions within a window around the CpG site: determining individual amounts of free DNA molecules having ends located at said positions, thereby determining a plurality of individual amounts; and The feature vector includes the respective quantities of the at least two positions and the respective quantities of the CpG sites.

137. The method of claim 129, wherein the first group of CpG sites is a plurality of CpG sites.

138. The method of claim 129, wherein the first amount is the amount of free DNA molecules with their termini at position 0, the method further comprising: determining a second amount of free DNA molecules having an end located within a first position within a window around any one of the first set of CpG sites, Wherein determining the proportional concentration comprises normalizing the first amount with the second amount to obtain a normalized first amount that is compared to the calibration value.

139. The method of claim 138, wherein the first position is +1 or -1 away from the CpG site.

140. The method of claim 138, wherein the window is at least -2 to +2 from the CpG site.

141. The method of claim 138, further comprising: For each of the at least two positions within the window: determining respective amounts of free DNA molecules having ends located at said positions, wherein said at least two positions include said first position; and The first amount of free DNA molecules ending at the CpG site is compared to the respective amounts of free DNA molecules ending at the position as part of determining the ratio concentration.

142. The method of claim 138, further comprising: For each of the at least two positions within the window: determining individual amounts of free DNA molecules having ends at said positions, thereby determining a plurality of individual amounts including said second amount, wherein said at least two positions include said first position; generating a feature vector including the respective quantities and the first quantity; and The feature vector is input into a machine learning model as part of determining the proportional concentration, wherein the machine learning model is trained using free DNA molecules from the one or more calibration samples having the known proportional concentration of DNA from the first tissue type.

143. The method of claim 142, wherein the one or more CpG sites of the first group in the reference genome are all hypomethylated for the first tissue type, and wherein the feature vector is a first feature vector, the method further comprising: generating a second feature vector using the one or more CpG sites of the second group that are all hypermethylated; and The second eigenvector and the first eigenvector are input into the machine learning model as part of determining the proportional concentration.

144. The method of claim 142, wherein the at least two positions include all positions within the window, the window being at least +4 to -4 from a CpG site.

145. The method of claim 142, wherein the machine learning model is support vector regression.

146. The method of claim 129, wherein the first amount is normalized.

147. A method as claimed in claim 146, wherein the normalization uses the number of free DNA molecules whose ends are located within the region including the CpG site.

148. The method of claim 146, wherein the normalization uses the number of free DNA molecules covering the CpG sites.

149. The method of claim 146, wherein the normalization uses the average or median depth of free DNA molecules in a region including CpG sites.

150. The method of claim 129, wherein the DNA ratio concentration is in the range of 30% or less.

151. The method of claim 129, wherein the first amount is determined for a free DNA molecule whose end is located at position 0 of any one of the CpG sites in the first group.

152. A method for measuring the proportional concentration of DNA from a first tissue type in a biological sample of a subject, the biological sample comprising cell-free DNA, the method comprising: Analyzing a plurality of cell-free DNA molecules from the biological sample of the subject, wherein analyzing the cell-free DNA molecules comprises: determining a genomic position in a reference genome corresponding to at least one end of the free DNA molecule; and Determining a terminal sequence motif of at least one end of the free DNA molecule, wherein the end of the free DNA molecule has a first position located at the outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position; identifying a first set of CpG sites in the reference genome that are all hypomethylated or hypermethylated for the first tissue type, wherein the first set of CpG sites comprises one or more CpG sites; A first quantity of the first set of one or more terminal sequence motifs is determined, wherein: The one or more terminal sequence motifs of the first group have a C at the first position and a G at the second position, or The one or more terminal sequence motifs of the first group have a C at the second position and a G at the third position; and The proportional concentration of DNA from the first tissue type in the biological sample is determined by comparing the first amount to a calibration value determined from one or more calibration samples having a known proportional concentration of DNA from the first tissue type.

153. A method for measuring 5hmC methylation of a region in a genome of a subject using free DNA molecules, the method comprising: Analyzing a plurality of cell-free DNA molecules from a biological sample of the subject, wherein analyzing the cell-free DNA molecules comprises: determining a terminal sequence motif of at least one end of the free DNA molecule, wherein the end of the free DNA molecule has a first position located at an outermost position, a second position immediately adjacent to the first position, and a third position immediately adjacent to the second position, wherein the plurality of free DNA molecules are located in the region of the reference genome; A first quantity of the first set of one or more terminal sequence motifs is determined, wherein: The one or more terminal sequence motifs of the first group have a C at the first position and a G at the second position, or The one or more terminal sequence motifs of the first group have a C at the second position and a G at the third position; and The classification of 5hmC methylation of the region in the genome of the subject is determined by comparing the first amount to a calibration value, wherein the calibration value is determined using free DNA molecules from one or more calibration samples and located at CpG sites with a known classification for 5hmC methylation levels.

154. The method of claim 153, wherein the first group of one or more terminal sequence motifs has a C at the first position and a G at the second position.

155. The method of claim 154, further comprising: determining a second amount of one or more terminal sequence motifs of a second group having a C at the second position and a G at the third position, Wherein determining the classification comprises normalizing the first quantity with the second quantity to obtain a normalized first quantity that is compared with the calibration value.

156. The method of claim 153, wherein the first group of one or more terminal sequence motifs has a C at the second position and a G at the third position.

157. The method of claim 153, further comprising: determining an expected amount of one or more terminal sequence motifs of the first group based on a reference sequence of the reference genome in the region, Wherein determining the classification comprises normalizing the first quantity with the expected quantity to obtain a normalized first quantity that is compared to the calibration value.

158. The method of claim 153, further comprising: determining an individual amount of each 3-mer terminal sequence motif having a C at the first position and a G at the second position, thereby determining a plurality of individual amounts; generating a feature vector including the quantities, wherein the quantities include the first quantity; and The feature vector is input into a machine learning model as part of determining the classification of the region in the genome of the subject, wherein the machine learning model is trained using free DNA molecules from the one or more calibration samples.

159. The method of any of the preceding claims, wherein analyzing the plurality of free DNA molecules uses sequencing.

160. The method of any of the preceding claims, wherein analyzing the plurality of free DNA molecules uses PCR.

161. The method of claim 160, wherein the PCR targets a sequence in a repetitive region.

162. The method of any of the preceding claims, wherein analyzing the plurality of free DNA molecules comprises analyzing at least 1,000 free DNA molecules.

163. A computer product comprising a non-transitory computer readable medium storing a plurality of instructions which, when executed, cause a computer system to perform the method of any of the preceding claims.

164. A system comprising: The computer product of claim 163, and One or more processors for executing instructions stored on the computer-readable medium.

165. A system comprising means for performing any of the above methods.

166. A system comprising one or more processors configured to perform any of the above methods.

167. A system comprising modules for respectively performing the steps of any one of the above methods.

Images