A method and device for predicting sample source based on methylation difference

Abstract

A method and device for predicting the source of a sample based on methylation differences, the method comprising a methylation state acquisition step, comprising acquiring the methylation state in the methylation sequencing data of the sample to be tested; a fragment analysis step, comprising using at least one selected from a control group baseline probability algorithm, an average methylation rate algorithm, and a spatial continuity algorithm to analyze the methylation state of the fragment in the sequencing data of the sample to be tested; a prediction step, comprising predicting that the sample to be tested is derived from a diseased individual or a control group individual, or predicting the probability that the sample to be tested is derived from a diseased individual based on the control group baseline data. The method effectively improves the accuracy of specific DNA (including but not limited to tissue-specific DNA, cell-specific DNA, tumor-specific DNA, or disease-specific DNA identification).

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a method and device for predicting the source of a sample based on methylation differences. Background Art

[0002] The identification of circulating cell-free tumor DNA (ctDNA) is increasingly recognized as an important diagnostic tool for non-invasive liquid biopsy that can be used to detect, diagnose, and monitor cancer. And DNA methylation is recognized to play an important role in regulating gene expression. DNA methylation profiles using whole genome methylation sequencing (e.g., whole genome bisulfite sequencing (WGBS), whole genome enzymatic sequencing (GM-seq)) can identify ctDNA molecular markers with disease / cancer specific methylation patterns. However, the art still needs improved methods for identifying specific methylation patterns with higher sensitivity and better specificity to detect, diagnose, and monitor diseases such as cancer. Summary of the invention

[0003] According to the first aspect, in one embodiment, a method for predicting the source of a sample based on methylation differences is provided, comprising:

[0004] The methylation status acquisition step includes acquiring the methylation status in the methylation sequencing data of the sample to be tested;

[0005] The fragment analysis step includes analyzing the methylation status of the fragments in the sequencing data of the sample to be tested using at least one selected from a control group baseline probability algorithm, an average methylation rate algorithm, and a spatial continuity algorithm;

[0006] The prediction step includes predicting whether the sample to be tested is derived from the diseased individual or the control group individual based on the baseline data of the control group, or predicting the probability that the sample to be tested is derived from the diseased individual.

[0007] According to the second aspect, in one embodiment, a device for predicting the source of a sample based on methylation differences is provided, comprising:

[0008] A methylation status acquisition module is used to obtain the methylation status in the methylation sequencing data of the sample to be tested;

[0009] A fragment analysis module, used to analyze the methylation status of fragments in the sequencing data of the sample to be tested using at least one selected from a control group baseline probability algorithm, an average methylation rate algorithm, and a spatial continuity algorithm;

[0010] The prediction module is used to predict whether the sample to be tested is derived from the diseased individual or the control group individual, or to predict the probability that the sample to be tested is derived from the diseased individual based on the baseline data of the control group.

[0011] According to the third aspect, in one embodiment, a device for predicting the source of a sample based on methylation differences is provided, comprising:

[0012] Memory, used to store programs;

[0013] A processor, configured to implement any one of the methods of the first aspect by executing a program stored in the memory.

[0014] According to the fourth aspect, in one embodiment, a computer-readable storage medium is provided, on which a program is stored, and the program can be executed by a processor to implement any method as described in the first aspect.

[0015] According to the method and device for predicting the source of a sample based on methylation differences of the above embodiment, the method effectively improves the accuracy of specific DNA recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a process for identifying ctDNA and building a model to identify malignant tumors in one embodiment;

[0017] Figure 2 The figure is a ROC curve diagram of the average methylation rate and spatial continuity (scm) prediction of 30 liver cancer patients and 30 control groups in one embodiment;

[0018] Figure 3 It is a ROC curve diagram of baseline probability filtering of the control group and filtering prediction of all algorithms for 30 liver cancer patients and 30 control groups in an embodiment. DETAILED DESCRIPTION

[0019] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. In the following embodiments, many detailed descriptions are intended to enable the present application to be better understood. However, those skilled in the art can effortlessly recognize that some of the features can be omitted in different situations, or can be replaced by other materials and methods. In some cases, some operations related to the present application are not shown or described in the specification, in order to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0020] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0021] The serial numbers assigned to the components in this article, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning.

[0022] Explanation of terms

[0023] As used herein, cell-free DNA (cfDNA) refers to DNA fragments from a sample (eg, plasma) that are released by apoptosis or death of cells, viruses, embryos, or cancer cells.

[0024] As used herein, circulating tumor DNA (ctDNA) refers to DNA fragments released by apoptosis or death of tumor cells or different types of cancer cells.

[0025] As used herein, CpG is an abbreviation for cytosine (C)-phosphate (p)-guanine (G).

[0026] According to the first aspect, in one embodiment, a method for predicting the source of a sample based on methylation differences is provided, comprising:

[0027] The methylation status acquisition step includes acquiring the methylation status in the methylation sequencing data of the sample to be tested;

[0028] The fragment analysis step includes analyzing the methylation status of the fragments in the sequencing data of the sample to be tested using at least one selected from a control group baseline probability algorithm, an average methylation rate algorithm, and a spatial continuity algorithm;

[0029] The prediction step includes predicting whether the sample to be tested is derived from the diseased individual or the control group individual based on the baseline data of the control group, or predicting the probability that the sample to be tested is derived from the diseased individual.

[0030] In one embodiment, in the prediction step, based on the relationship between the control group baseline probability of the fragment in the sample to be tested and a preset threshold, it is predicted whether the fragment is an abnormal fragment different from the control group individual.

[0031] In one embodiment, in the prediction step, if the control group baseline probability of the fragment in the sample to be tested is lower than (i.e., smaller than) a preset threshold, the fragment is predicted to be a fragment different from the control group individual; otherwise, the fragment is predicted to be a fragment derived from the control group individual.

[0032] In one embodiment, in the prediction step, the control group baseline data includes the average methylation rate corresponding to each 2, 3 or N adjacent CpG sites in the target region of the control group samples, where N is a natural number greater than 3.

[0033] In one embodiment, in the prediction step, the control group baseline is a baseline established based on a certain number of sample methylation sequencing data from individuals in the control group.

[0034] In one embodiment, if the purpose is to identify ctDNA from cfDNA of the test sample, the blood cfDNA methylation sequencing data of the control group individuals can be used to construct the control group baseline.

[0035] In one embodiment, if the purpose is to identify tissue-specific, cell-specific or disease-specific DNA from the genomic DNA of the test sample, the methylation sequencing data of the genomic DNA of the control group individuals (including but not limited to genomic DNA derived from tissues) can be used to construct a control group baseline.

[0036] In one embodiment, the fragment analysis step includes using a control group baseline probability algorithm to analyze the methylation status of the fragments in the sequencing data of the sample to be tested.

[0037] In one embodiment, the fragment analysis step includes using a) a control group baseline probability algorithm, and b) at least one of an average methylation rate algorithm and a spatial continuity algorithm to analyze the methylation status of the fragments in the sequencing data of the sample to be tested.

[0038] In one embodiment, in the fragment analysis step, the calculation formula of the control group baseline probability is as follows:

[0039] P(frag)=P(S0)×P(S01|S0)×P(S12|S1)...×P(S(n-1)n|S(n-1)) (1)

[0040] In formula (1), P(frag) is the control group baseline probability of the fragment, and P(S0) is the probability of the corresponding methylation state appearing at the first group of CpG sites in the fragment (for example, the probability of the first CpG site, or the first two, or the first n CpG sites appearing at the corresponding methylation state);

[0041] P(S(n-1)n│S(n-1)) is the probability that the corresponding methylation state appears in the nth group of CpG sites in the fragment when the corresponding methylation state appears in the n-1th group of CpG sites in the fragment.

[0042] In one embodiment, in the fragment analysis step, the calculation formula of spatial continuity (scm) is as follows:

[0043]

[0044] In formula (2), l is the total number of CpG sites covered by the fragment, and N(i) is the number of sub-fragments with i consecutive CpG sites with the same specific methylation state from the starting CpG site to the ending CpG site of the fragment. i is the length weight of the sub-segment.

[0045] In one embodiment, in the fragment analysis step, the average methylation rate of the fragment is calculated based on the ratio of the number of methylated CpG sites in the fragment of the sequencing data of the sample to be tested to the total number of CpG sites.

[0046] In one embodiment, after analyzing the methylation information of the fragments using at least one of the algorithms selected from the group consisting of the average methylation rate of the fragments, spatial continuity (SCM), and the baseline probability algorithm of the control group in the fragment analysis step, in the prediction step, based on at least one of the following information in the sequencing data of the sample to be tested: 1) the average methylation rate of all fragments in the target region, 2) the spatial continuity of all fragments in the target region, and 3) the proportion of specific DNA fragments in the target region; it is predicted that the sample to be tested is derived from a diseased individual or a control group individual, or the probability that the sample to be tested is derived from a diseased individual is predicted.

[0047] In one embodiment, the specific DNA includes but is not limited to tissue-specific DNA, cell-specific DNA or disease-specific DNA;

[0048] In one embodiment, the specific DNA includes but is not limited to ctDNA.

[0049] In one embodiment, the specific DNA proportion refers to the proportion of specific DNA in all fragments of sequencing data.

[0050] In one embodiment, the ctDNA proportion refers to the proportion of ctDNA in all fragments of sequencing data.

[0051] In one embodiment, in the prediction step, a model is used to predict whether the sample to be tested is derived from a diseased individual or a control group individual, or to predict the probability that the sample to be tested is derived from a diseased individual.

[0052] In one embodiment, the model includes but is not limited to a machine learning model or a probability distribution model.

[0053] In one embodiment, the machine learning model includes but is not limited to a support vector machine model, a random forest model, a logistic regression model, a deep learning model, an extreme gradient boosting model, or a gradient boosting tree model.

[0054] In one embodiment, the probability distribution model includes but is not limited to a Bernoulli mixture distribution model, a Bayesian model, or a Beta distribution model.

[0055] In one embodiment, a machine learning model is used to learn the training set feature weights of the sequencing data of diseased samples and the sequencing data of control samples to identify diseased samples. Diseased samples refer to samples from diseased individuals.

[0056] In one embodiment, the probability value of the sample to be tested coming from a diseased individual is determined based on the input features of the sample to be tested, and the probability value is compared with a preset probability threshold to determine whether the sample to be tested comes from a diseased individual.

[0057] In one embodiment, in the analysis step, after analyzing the methylation status of the fragments using at least one of the average methylation rate algorithm of the fragments, the spatial continuity (SCM) algorithm, and the control group baseline probability algorithm, in the prediction step, the average methylation rate of all fragments in the target region, the average value of at least one of the spatial continuity (SCM) indicators are used as features, or the identified specific DNA ratio is used as a feature to construct a prediction model to identify whether the individual to which the sample to be tested belongs has a disease. Prediction models include but are not limited to machine learning models.

[0058] In one embodiment, the disease includes but is not limited to malignant tumors or other diseases.

[0059] In one embodiment, the specific DNA includes but is not limited to tissue-specific DNA, cell-specific DNA, and disease-specific DNA.

[0060] In one embodiment, the specific DNA includes but is not limited to ctDNA.

[0061] In one embodiment, in the prediction step, the average methylation rate of all fragments in the target region and the average value of the spatial continuity index are used as features to construct a machine learning model to predict the disease suffered by the individual to whom the sample to be tested belongs, or the probability of the disease suffered by the individual.

[0062] In one embodiment, in the prediction step, the average methylation rate and spatial continuity of each retained abnormal fragment in the target region are calculated, and they are preprocessed into a two-dimensional matrix with the horizontal axis being the average methylation rate of each target region or the methylation level represented by the spatial continuity, and the vertical axis being each sample, to construct a prediction model.

[0063] In one embodiment, in the prediction step, the identified specific DNA proportion is used as a feature to construct a machine learning model to predict the disease suffered by the individual to whom the sample to be tested belongs, or the probability of the disease suffered by the individual.

[0064] In one embodiment, in the prediction step, if the target region is a disease hypomethylated region, only fragments whose spatial continuity is greater than a preset threshold or whose average methylation rate is less than a preset threshold are retained, and the model is constructed using the proportion of the co-methylated fragment counts of the disease-specific hypomethylation in the methylated fragment counts of all quality control-qualified fragments. The model is preprocessed into a two-dimensional matrix with the horizontal axis being the specific DNA proportion of each target region and the vertical axis being each sample to construct a prediction model.

[0065] In one embodiment, the specific DNA includes but is not limited to tissue-specific DNA, cell-specific DNA, and disease-specific DNA.

[0066] In one embodiment, the specific DNA includes but is not limited to ctDNA.

[0067] In one embodiment, the disease includes but is not limited to cancer or other diseases.

[0068] In one embodiment, the cancer includes liver cancer. This is only an example, and the cancer types applicable to the present invention are not limited.

[0069] In one embodiment, in the prediction step, the disease suffered by the patient includes but is not limited to cancer or other diseases.

[0070] In one embodiment, the cancer includes, but is not limited to, lymphoma, blastoma, sarcoma, leukemia, lymphoid malignancies, squamous cell carcinoma (e.g., epithelial squamous cell carcinoma), skin cancer, melanoma, lung cancer, peritoneal cancer, gastric cancer, pancreatic cancer (e.g., pancreatic ductal adenocarcinoma), cervical cancer, ovarian cancer (e.g., high-grade serous ovarian cancer), liver cancer (e.g., hepatocellular carcinoma (HCC)), liver cancer, bladder cancer (e.g., urothelial bladder cancer), testicular (germ cell tumor) cancer, breast cancer (e.g., HER2-positive, HER2-negative, and triple-negative breast cancer), brain cancer (e.g., astrocytoma, glioma (e.g., glioblastoma), colon cancer, rectal cancer, colorectal cancer, endometrial or uterine cancer, salivary gland cancer, kidney cancer (e.g., renal cell carcinoma, Wilms' tumor, or Wilms' tumor), prostate cancer, vulvar cancer, thyroid cancer, anal cancer, penile cancer, head and neck cancer, esophageal cancer, nasopharyngeal carcinoma (NPC), retinoblastoma, theca cell tumor, androma, hematological malignancies.

[0071] In one embodiment, the lung cancer is but not limited to small cell lung cancer, non-small cell lung cancer ("NSCLC"), lung adenocarcinoma, and lung squamous cell carcinoma.

[0072] In one embodiment, the gastric cancer comprises gastrointestinal cancer.

[0073] In one embodiment, other diseases include but are not limited to adverse reactions after organ transplantation, infection by pathogens such as bacteria, viruses, fungi, parasites, and chronic diseases such as hepatitis, chronic gastroenteritis, diabetes, chronic pharyngitis, cardiovascular and cerebrovascular diseases.

[0074] In one embodiment, in the methylation status acquisition step, the sample to be tested includes but is not limited to a body fluid sample or a tissue sample.

[0075] In one embodiment, the body fluid sample includes but is not limited to plasma from blood, urine, saliva, bronchoalveolar lavage fluid or cerebrospinal fluid.

[0076] In one embodiment, in the methylation status acquisition step, the methylation sequencing includes but is not limited to whole genome methylation sequencing (WGBS), degenerate methylation sequencing (RRBS) or whole genome enzyme conversion sequencing.

[0077] In one embodiment, in the methylation status acquisition step, the methylation sequencing is to convert the unmodified cytosine (C) in the nucleic acid fragment into a base that does not bind to guanine (e.g., uracil base U) or to convert the modified cytosine into a third modification, such as a carboxylated cytosine base (C), by an enzyme, which is further identified as a thymine base (T).

[0078] In one embodiment, in the step of obtaining the methylation status, the methylation sequencing is performed by treating the nucleic acid fragments through a non-enzymatic (eg, bisulfite or bisulfate treatment) or an enzymatic reaction.

[0079] In one embodiment, the nucleic acid fragments include but are not limited to cfDNA or gDNA (genomic DNA).

[0080] In one embodiment, in the methylation status acquisition step, the methylation sequencing data is data aligned to a reference genome.

[0081] In one embodiment, the reference genome includes but is not limited to GRCH37, b37, hs37d5 (b37+decoy), hg19, GRCH38 (hg38), etc. In one embodiment, hg19 can be downloaded from UCSC (http: / / genome.ucsc.edu / ), and GRCH38 can be downloaded from NCBI (https: / / www.ncbi.nlm.nih.gov / ). In one embodiment, if the sample is taken from a human body, the sequencing data is aligned to the human reference genome, and if the sample is taken from an animal, the sequencing data is aligned to the reference genome of the corresponding animal species.

[0082] In one embodiment, the prediction step includes obtaining multiple nucleic acid fragments from individuals in the control group, identifying the methylation status of each known CpG site based on the reference genome, and counting the corresponding methylated and unmethylated fragment counts to obtain the average methylation rate, and then estimating the corresponding parameters of the Bernoulli probability distribution or the beta probability distribution, and finally multiplying the result to obtain the baseline probability of the control group.

[0083] In one embodiment, in the prediction step, the nucleic acid fragments obtained from the control group individuals include but are not limited to cfDNA or gDNA (genomic DNA).

[0084] In one embodiment, in the methylation status acquisition step, a method for extracting information of co-methylation (linked methylation) fragments is used to acquire the methylation status in the methylation sequencing data of the sample to be tested.

[0085] In one embodiment, the methylation status acquisition step includes acquiring the methylation status of a fragment in a target region of the genomic DNA of the sample to be tested.

[0086] In one embodiment, the methylation status acquisition step includes acquiring the methylation status of cfDNA fragments in the target region in the whole genome methylation sequencing data of the cfDNA to be tested.

[0087] In one embodiment, the target region can be determined based on the target region investigated by the previous literature or the inventor, and then the signal-to-noise ratio of the region can be verified in the real methylation sequencing data to determine the region that can actually provide a signal, that is, the target region. The determination of the target region can be carried out by conventional methods. For example, the target region can be determined by referring to the Chinese patent "A method and device for screening differentially methylated regions" with publication number CN114171115A.

[0088] In one embodiment, in the methylation status acquisition step, the sample to be tested is derived from a human or an animal, preferably a human.

[0089] According to the second aspect, in one embodiment, a device for predicting the source of a sample based on methylation differences is provided, comprising:

[0090] A methylation status acquisition module is used to obtain the methylation status in the methylation sequencing data of the sample to be tested;

[0091] A fragment analysis module, used to analyze the methylation status of fragments in the sequencing data of the sample to be tested using at least one selected from a control group baseline probability algorithm, an average methylation rate algorithm, and a spatial continuity algorithm;

[0092] The prediction module is used to predict whether the sample to be tested is derived from the diseased individual or the control group individual, or to predict the probability that the sample to be tested is derived from the diseased individual based on the baseline data of the control group.

[0093] According to the third aspect, in one embodiment, a device for predicting the source of a sample based on methylation differences is provided, comprising:

[0094] Memory, used to store programs;

[0095] A processor, configured to implement any one of the methods of the first aspect by executing a program stored in the memory.

[0096] According to the fourth aspect, in one embodiment, a computer-readable storage medium is provided, on which a program is stored, and the program can be executed by a processor to implement any method as described in the first aspect.

[0097] It should be noted that the sequencing data in the methylation status acquisition step is the data of the in vitro sample, therefore, it is not based on a living human or animal body; and, in the prediction step, the predicted result of the sample to be tested is only an intermediate result, which is used as a reference for subsequent disease diagnosis, and belongs to intermediate reference information, not the final diagnosis result. In practical applications, after using the method of the present invention to predict that the sample to be tested is derived from a diseased individual or a control group individual, or predicting the probability that the sample to be tested is derived from a diseased individual, it is also necessary to combine the subject's current subjective symptoms, past medical history, family genetic history and other information to obtain the final diagnosis result or the control group condition. Therefore, the prediction result of the prediction step according to the present invention alone cannot obtain a diagnosis result in the sense of the patent law. Therefore, the technical solution of the present invention does not belong to a method for diagnosing a disease, let alone a method for treating a disease.

[0098] In one embodiment, the present invention provides an algorithm for identifying ctDNA fragments in plasma or other body fluid samples and a method for further constructing an early cancer identification model, the method comprising: (1) using a method for extracting information of co-methylated (linked methylated) fragments to obtain the methylation status of cfDNA fragments in a target region in whole genome methylation sequencing data of tumor plasma or other body fluid samples and control group plasma or other body fluid samples; (2) using at least one algorithm indicator selected from the group consisting of average methylation rate, spatial continuity (linked methylation index scm), and control group baseline probability algorithm to process the methylation status of the fragments; (3) screening out fragments with high methylation levels in tumor plasma samples and control group plasma samples that are derived from tumor cells, the fragments being ctDNA; and (4) constructing an early cancer identification model based on the methylation level and content of ctDNA.

[0099] It is very important to detect cancer in subjects as early as possible, and early treatment often leads to better prognosis and longer survival. Cell-free DNA (cfDNA) includes the stitching and sequencing of circulating cell-free tumor DNA (ctDNA) fragments and the determination and analysis of the methylation status of various dinucleotides of cytosine and guanine (called CpG sites) in the fragments, which can help predict whether the subject has cancer. In one embodiment, the present invention provides a method for processing the methylation status of fragments using at least one algorithm indicator of the average methylation rate, spatial continuity (scm), and control group baseline probability algorithm, and can more sensitively identify ctDNA.

[0100] The control group baseline probability algorithm can assist in determining fragments with abnormal methylation patterns. Methylation fragments with a fixed pattern that appear frequently and repeatedly in control group individuals may not necessarily provide highly recognizable signals for distinguishing cancer status. Therefore, a control group baseline established by a certain number of control group individuals can be used as a reference to calculate the probability of occurrence of fragments with a specific methylation pattern in the control group population, that is, the control group baseline probability. If the control group baseline probability of a fragment is lower than a certain threshold, the fragment is considered to be a fragment different from that of the control group. The control group baseline probability algorithm formula is as follows:

[0101] P(frag)=P(S0)×P(S01|S0)×P(S12|S1)...×P(S(n-1)n|S(n-1)) (1)

[0102] In formula (1), P(frag) is the baseline probability of the control group of the fragment, P(S0) is the probability of the first group of CpG sites in the fragment having the corresponding methylation state (for example, the probability of the first CpG site, or the first two, or the first n CpG sites having the corresponding methylation state). P(S(n-1)n│S(n-1)) is the probability of the corresponding methylation state of the nth group of CpG sites in the fragment, based on the corresponding methylation state of the n-1th group of CpG sites in the fragment.

[0103] Because the methylation state of a CpG site is affected by the methylation state of an adjacent CpG site, the control group baseline probability chain algorithm can effectively calculate the probability of occurrence of such a given methylation state fragment. In one embodiment, the present invention assumes that the methylation state of the current CpG site is affected by the methylation state of the previous group of CpG sites (e.g., affected by the methylation state of the first 1, first 2, first 3, or first n CpG sites).

[0104] In one embodiment, multiple cfDNA fragments are obtained from a control group subject (control group), the methylation status of each known CpG site is identified according to the reference genome, and the corresponding methylated and unmethylated fragment counts are counted to obtain the average methylation rate to estimate the corresponding parameters of the Bernoulli probability distribution or the beta probability distribution. For a single fragment of the test sample, the control group baseline probability is calculated using the above formula (1), and this probability is alternative to the probability of listing the actual occurrence of this fragment. Therefore, if the control group baseline probability is lower than a preset threshold, such as 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, it is considered that the probability of this fragment appearing in the control group population is lower than 0.01%, 0.05%, 0.1%, 0.5%, 1%, 5%, 10%, and it is an abnormal fragment different from the control group. The present invention also provides two indicators, spatial continuity (scm) and average methylation rate of fragments, for more accurate identification of ctDNA.

[0105] Because of the specificity of gene expression in different cells, methylation also fully demonstrates its cell-specific / tissue-specific characteristics. Moreover, the methylation state of CpG sites is affected by the methylation state of adjacent CpG sites, and co-methylation is prone to occur. This cell-specific co-methylation characteristic can eliminate sequencing or experimental conversion errors and more sensitively identify ctDNA. In one embodiment, the spatial continuity (scm) index provided by the present invention can have higher recognition on the basis of a certain fault tolerance rate compared to the conventionally used regional average methylation rate. If the current fragment is located in a specific low-methylation region, the spatial continuity (scm) index corresponding to all CpG sites in the fragment being unmethylated is the maximum value of 1, and the spatial continuity (scm) index corresponding to the fragment having no or only one CpG site being unmethylated is the minimum value of 0. The more continuous unmethylated CpG sites in the same fragment, the stronger the co-methylation and the higher the spatial continuity index.

[0106] Hypermethylation means that the methylation level of the disease is higher than that of the control group.

[0107] Hypomethylation means that the methylation level of the disease is lower than that of the control group.

[0108] The algorithm of spatial continuity (scm) is as follows:

[0109]

[0110] In formula (2), l is the total number of CpG sites covered by the fragment, and N(i) is the number of sub-fragments with i consecutive CpG sites with the same specific methylation state from the starting CpG site to the ending CpG site of the fragment. i is the length weight of the sub-segment, which can be i.

[0111] The average methylation rate of the fragment can be calculated by the ratio of the number of methylated CpG sites in the fragment to the total number of CpG sites.

[0112] In one embodiment, if a specific region of the genome shows a specific low methylation trend in liver cancer cells, and other various cell types including but not limited to red blood cells, immune cells, heart, lung, kidney, liver and other normal tissue cells show a specific high methylation trend. The higher the spatial continuity (scm), the more continuous unmethylated CpG sites there are in this region, the lower the average methylation rate of the fragment, and the more likely this sample is to come from liver cancer.

[0113] In one embodiment, after processing the methylation state of the fragment using at least one algorithm index in the average methylation rate of the fragment, spatial continuity (scm), and control group baseline probability algorithm, the average methylation rate of all fragments in the target area and the average value of the spatial continuity (scm) index can be used as features, or the identified ctDNA proportion is used as a feature to construct a machine learning model to identify whether the individual to which the sample to be tested belongs suffers from a malignant tumor or other disease. The machine learning model includes but is not limited to a support vector machine model, a random forest model, a logistic regression model, a deep learning model, an extreme gradient boosting model (XGBoost), and a gradient boosting tree model (GBM). The model is preferentially constructed using the train function in the caret software of the R language. The model will score each sample, and when the score meets the threshold, the sample is identified as a malignant tumor or a control group.

[0114] In one embodiment, based on the target region previously investigated in the literature or by the inventors, the signal-to-noise ratio of the region is verified in the real methylation sequencing data, and the region that can actually provide a signal is determined to be the target region.

[0115] Cancers that can be detected using the methods, systems and classifiers of the present invention include, but are not limited to, malignant tumors, and specifically may include, but are not limited to, lymphomas, blastomas, sarcomas, leukemias or lymphoid malignancies. More specific examples of such cancers include, but are not limited to, squamous cell carcinomas (e.g., epithelial squamous cell carcinomas), skin cancers, melanomas, lung cancers, including small cell lung cancers, non-small cell lung cancers ("NSCLC"), lung adenocarcinomas and lung squamous cell carcinomas, peritoneal cancers, gastric cancers or gastric cancers including gastrointestinal cancers, pancreatic cancers (e.g., pancreatic ductal adenocarcinomas), cervical cancers, ovarian cancers (e.g., high-grade serous ovarian cancers), liver cancers (e.g., hepatocellular carcinoma (HCC)), liver cancers, bladder cancers (e.g., urothelial bladder cancers), testicular (germ cell tumor) cancers, breast cancers (e.g., HER2-positive, HER2-negative and triple-negative breast cancers). Cancer, for example, astrocytoma, glioma (e.g., glioblastoma), colon cancer, rectal cancer, colorectal cancer, endometrial or uterine cancer, salivary gland cancer, kidney cancer (e.g., renal cell carcinoma, Wilms' tumor, or Wilms' tumor), prostate cancer, vulvar cancer, thyroid cancer, anal cancer, penile cancer, head and neck cancer, esophageal cancer, and nasopharyngeal carcinoma (NPC). Other examples of cancer include, but are not limited to, retinoblastoma, theca cell tumor, androgenic tumor, hematological malignancies, including, but not limited to, non-Hodgkin's lymphoma (NHL), multiple myeloma, and acute hematological malignancies.

[0116] Other examples of cancer may also include, but are not limited to, endometriosis, fibrosarcoma, choriocarcinoma, laryngeal cancer, Kaposi's sarcoma, schwannoma, neuroblastoma, rhabdomyosarcoma, osteogenic sarcoma, leiomyosarcoma, and urinary tract cancer.

[0117] In one embodiment, the sample types detected by the method, system and classifier of the present invention are not limited to plasma samples, but can also detect other body fluid samples such as urine, saliva, alveolar lavage fluid, cerebrospinal fluid, etc.

[0118] In one embodiment, the present invention uses whole genome methylation sequencing data of cfDNA from body fluid samples such as plasma to identify and detect ctDNA of early cancer. Based on this, a machine learning model is constructed to predict whether a single sample has a risk of cancer.

[0119] Example 1

[0120] Sample processing

[0121] Figure 1 Shown is a schematic diagram of the process of identifying ctDNA and building a model to identify malignant tumors in this embodiment.

[0122] First, the plasma samples of the subjects are collected, and cell-free DNA (cfDNA) is extracted. Methylation sequencing is performed on the extracted DNA samples, including but not limited to whole genome methylation sequencing (WGBS), or degenerate methylation sequencing (RRBS), or whole genome enzyme conversion sequencing (GM-seq). Methylation sequencing is the use of non-enzymatic (such as bisulfite or bisulfate treatment) or enzymatic reaction to treat cfDNA, converting unmodified cytosine (C) into bases that do not bind to guanine (such as uracil base U) or converting modified cytosine into a third modification by enzyme. The sequence reads are compared with the reference genome (hg19 in this example) to determine the location and methylation status of each methylation site of the cfDNA fragment. If it is a double-end comparison, read1 and read2 are additionally stitched according to established rules to obtain the methylation status of the entire fragment. In this example, 30 healthy people (control group) blood cfDNA and 30 liver cancer patients' blood cfDNA were used for whole genome enzyme conversion sequencing (GM-seq). 70% of the data is used as a training set for cross-validation, and the remaining 30% is used to test the performance of the overall machine learning model.

[0123] Abnormal fragment identification / ctDNA preliminary filtering

[0124] Determine the methylation characteristics of the DNA region and the algorithm, and screen out the region that best represents the difference between liver malignancies and healthy controls, which are referred to as target regions hereinafter. This embodiment refers to the Chinese patent "A method for screening differential methylation regions and its device" with publication number CN114171115A, and determines the target region (i.e., the target region). After deduplication of this target region fragment, according to the quality control of the information such as comparison quality and length, a control group baseline is established using the fragments of multiple control groups or control group subjects, that is, the methylation state counts of each CpG site in the target region are counted, including methylation and unmethylation, so as to calculate the control group baseline probability of each fragment of the test sample later. The control group baseline data will be stored and called with the number of fragments corresponding to the methylation state of each 2, 3 or N adjacent CpG sites in the target region (the present embodiment is to call the number of fragments of each 2 adjacent CpG sites, different target regions, different thresholds), and the average methylation rate is obtained from the number of fragments to estimate the parameters of the beta distribution. The actual calculation of the control group baseline probability of each fragment is the result of the multiplication of the beta distribution parameters of the corresponding CpG sites. If the control group baseline probability calculated for a certain fragment is less than a given threshold, such as 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, then the fragment is retained for the next analysis; if it is higher than the threshold, it is eliminated.

[0125] Model building

[0126] This embodiment provides a binary classifier that uses a machine learning model to learn the weights of training set features of cancer and non-cancer to identify malignant tumors. The classifier divides the original data into multiple groups, performs iterative batch training and verification, and confirms the possibility or score of the sample coming from a cancer subject based on the sample input features. In this embodiment, the probability score is compared with a threshold probability to ultimately determine whether the subject has cancer.

[0127] In this embodiment, the average methylation rate of all fragments in the target area and the average value of the spatial continuity (scm) index can be used as features to construct a machine learning model to identify malignant tumors or other diseases. The average methylation rate and spatial continuity (scm) of each abnormal fragment retained in the target area are calculated, and the abscissa is the methylation level represented by the average methylation rate or spatial continuity (scm) of each region, and the ordinate is a two-dimensional matrix for each sample. The train function in the caret software package of R language is used to build a malignant tumor recognition model, which is a support vector machine model (SVM). In the R language environment, the caret module is loaded, and the ctControl parameter is set to 5x cross validation, ctrl=trainControl(method="cv", number=5, summaryFunction=twoclassSummary). The SVM model of this embodiment is model_svm=train(x,y,method=“svmLinear2”,metric=“ROC”,trControl=ctrl), where x is a two-dimensional matrix of methylation levels represented by the average methylation rate or spatial continuity, and y is the category information of the samples used for modeling, liver cancer is cancer, and the control group is health.

[0128] The specific prediction method for the sample to be tested is as follows: calculate the average methylation rate and spatial continuity (SCM) of each retained abnormal fragment in the target area of ​​the sample to be tested, and preprocess it into a two-dimensional matrix with the horizontal axis being the average methylation rate or spatial continuity (SCM) of each region representing the methylation level, and the vertical axis being the sample label. Input this matrix into the trained support vector machine model, and output the prediction of the corresponding sample as the probability of cancer origin. If the predicted probability of cancer origin is greater than the threshold obtained by training, it is determined that the sample to be tested comes from a cancer patient.

[0129] The performance comparison results of the above two methylation level indicators in samples are shown in Table 1 and Figure 2 shown. Figure 2 This is the ROC curve of the average methylation rate and spatial continuity (scm) prediction of 30 liver cancer patients and 30 control groups.

[0130] Table 1

[0131]

[0132]

[0133] Accuracy refers to the proportion of samples whose predicted labels are consistent with the true labels out of the total number of samples used.

[0134] From Table 1 and Figure 2 It can be seen that the AUCs of the two algorithm indicators are both greater than 0.9, showing good predictive ability, among which spatial continuity has better specificity and higher accuracy.

[0135] In this embodiment, the identified ctDNA ratio can also be used as a feature to construct a machine learning model to identify malignant tumors or other diseases. If the target area is a hypomethylated area of ​​liver cancer, only the spatial continuity (scm) is greater than the established threshold (for example, the threshold is 0.1, 0.2, 0.5, 0.8), and the fragments with an average methylation rate less than the established threshold (for example, the threshold is 0.2, 0.33, 0.4) are retained. The fragments that have been filtered twice are more likely to be co-methylated fragments with specific hypomethylation of liver cancer. The model is constructed using the proportion of the co-methylated fragment counts of liver cancer-specific hypomethylation in the methylated fragment counts of all qualified quality control. It is preprocessed into a two-dimensional matrix with the horizontal axis as the ctDNA ratio of each region and the vertical axis as each sample. The train function in the caret software package of R language is used to construct a malignant tumor recognition model, which is a support vector machine model. In the R language environment, load the caret module, set the ctControl parameter to 5x cross validation, ctrl = trainControl (method = "cv", number = 5, summaryFunction = twoclassSummary). The SVM model of this embodiment is model_svm = train (x, y, method = "svmLinear2", metric = "ROC", trControl = ctrl), where x is the average methylation rate or a two-dimensional matrix of methylation levels represented by spatial continuity, y is the category information of the sample used for modeling, liver cancer is cancer, and the control group is health. The performance comparison of the above two ctDNA level indicators in the samples of the present invention is shown in Table 2 and Figure 3 shown. Figure 3 The ROC curves of baseline probability filtering and all algorithm filtering prediction for 30 liver cancer patients and 30 control groups are shown.

[0136] Table 2

[0137]

[0138] As can be seen from Table 2, when the AUC values ​​are close (the AUC values ​​differ by less than 1%, which can be considered to be basically the same), using the three indicators at the same time can achieve higher prediction accuracy.

[0139] From Table 2 and Figure 3 It can be seen that the AUC of the two algorithm indicators are both greater than 0.95, showing better predictive ability than the previous method. Among them, the ctDNA identified by filtering all three algorithm indicators, namely, the baseline probability of the control group, the fragment methylation rate, and the spatial continuity, has better specificity, higher sensitivity, and higher accuracy.

[0140] In one embodiment, the spatial continuity algorithm index can provide better differentiation of early malignant tumors than the average methylation rate, and ctDNA can also process the methylation status of the fragment by using at least one algorithm index of the average methylation rate, spatial continuity (scm), and control group baseline probability. In one embodiment, three indicators can be used at the same time, which can identify ctDNA more sensitively and specifically than using one of the indicators.

[0141] In one embodiment, the methylation status of the target region of the genomic DNA of the tumor plasma sample and the control plasma sample is obtained. The information extraction method of the co-methylated (linked methylation) fragment of cfDNA in plasma is used, that is, the complete methylation signal is extracted from the reads in BS or non-BS sequencing, and the ctDNA information is sensitively obtained using methods such as spatial continuity (scm) and control group baseline probability. There is currently no method for stitching fragments based on GM-seq data to read the methylation status of each fragment.

[0142] In one embodiment, at least one algorithmic indicator selected from the group consisting of average methylation rate, spatial continuity (SCM), and control group baseline probability is used to process the methylation status of the fragment. These three indicators are calculated in units of fragments, rather than the average value within the region, which is more sensitive. And there is currently no program that combines these indicators for combined use.

[0143] In one embodiment, spatial continuity (scm) is an indicator for measuring the continuity of methylation status within a fragment, which is more sensitive than indicators such as average methylation rate.

[0144] In one embodiment, the fragments whose methylation levels meet the positive fragment determination in the tumor plasma samples and the control plasma samples are screened out, and the fragments are ctDNA. The control blood is used as the background and the tumor blood is used as the positive reference to improve the signal-to-noise ratio, providing a more accurate method for locating ctDNA.

[0145] In one embodiment, an early cancer identification model is constructed based on the methylation level and content of ctDNA. One of the machine models such as support vector machine, logistic regression, random forest, etc. is used, and the ctDNA fragment count / average methylation rate / average spatial continuity of the sample in each target area is used as input to determine early cancer.

[0146] In one embodiment, the specific DNA identified by the present invention is not limited to ctDNA, but also includes but is not limited to tissue-specific DNA, cell-specific DNA or various disease-specific DNA. The present invention can identify various specific DNA based on methylation status.

[0147] In one embodiment, the algorithm of the present invention for using the control group baseline probability calculation to identify abnormal fragments adopts a calculation method that assumes that the methylation status of the entire fragment is known and consistent. At the same time, special cases such as inconsistent methylation status of two read comparisons, a gap in the middle that cannot be completely stitched, or various mutations (snp, indel) may also be considered. These special cases can also be included in the calculation of the control group baseline through deformation.

[0148] The current example only provides a specific implementation method for early cancer screening. In one embodiment, this method can also be used to identify tissue-specific / cell-specific / disease-specific fragments, and can further be used for chronic or acute organ damage, or transplant monitoring, metastasis detection and other directions. In the case of a clear positive fragment methylation pattern in the target area, this method provides multiple dimensions of fragment recognition algorithm indicators. In one embodiment, the tissue-specific / cell-specific / disease-specific fragment can be a specific cfDNA fragment.

[0149] The biological interpretability of the deep learning models used in the prior art is poor. If a cfDNA is judged by the deep learning model to be derived from cancer, but the model cannot explain the biological principle of the judgment. In one embodiment, the indicators used in the present invention, such as spatial continuity (scm) and control group baseline probability, have certain biological interpretability. Spatial continuity (scm) characterizes that the methylation state of a CpG site is affected by the methylation state of adjacent CpG sites. The greater the spatial continuity (scm), the stronger the co-methylation. The control group baseline probability is the probability of each test fragment appearing in the control group population based on the methylation fragment pattern in the control group population, and this probability is highly correlated with the frequency of occurrence of the real fragment.

[0150] In one embodiment, gene expression is related to methylation, and the average methylation rate can roughly measure the methylation status in a region, but it is not as good as spatial continuity (spatial continuity also takes into account the co-methylation status on this basis).

[0151] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.

[0152] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.

Claims

1. A method for predicting the source of a sample based on methylation differences, comprising: The methylation status acquisition step includes acquiring the methylation status in the methylation sequencing data of the sample to be tested; The fragment analysis step includes using a control group baseline probability algorithm, an average methylation rate algorithm, and a spatial continuity algorithm to analyze the methylation status of the fragments in the sequencing data of the sample to be tested; A prediction step, comprising predicting whether the sample to be tested is derived from a diseased individual or a control group individual, or predicting the probability that the sample to be tested is derived from a diseased individual, based on the baseline data of the control group; In the prediction step, based on the relationship between the control group baseline probability value of the fragment in the sample to be tested and the preset threshold, it is predicted whether the fragment is an abnormal fragment different from the control group individual; In the prediction step, if the control group baseline probability value of the fragment in the sample to be tested is lower than a preset threshold, the fragment is predicted to be a fragment different from the control group individual; otherwise, the fragment is predicted to be a fragment derived from the control group individual; In the prediction step, the control group baseline data includes the average methylation rate corresponding to every 2, 3 or N adjacent CpG sites in the region, where N is a natural number greater than 3; In the prediction step, the control group baseline is a baseline established based on sample methylation sequencing data from individuals in the control group; In the fragment analysis step, the baseline probability of the control group is calculated as follows: （1） In formula (1), P(frag) is the control group baseline probability of the fragment, and P(S0) is the probability of the corresponding methylation state appearing at the first group of CpG sites in the fragment; P(S(n-1)n│S(n-1)) is the probability that the corresponding methylation state appears in the nth group of CpG sites in the fragment when the corresponding methylation state appears in the n-1th group of CpG sites in the fragment; In the fragment analysis step, the calculation formula for spatial continuity is as follows: （2） In formula (2), l is the total number of CpG sites covered by the fragment, N(i) From the start CpG site to the end CpG site of the fragment, it appears continuously i The number of sub-fragments with consistent specific methylation status at each CpG site, W i is the length weight of the sub-segment.

2. The method according to claim 1, characterized in that In the fragment analysis step, the average methylation rate of the fragment is calculated based on the ratio of the number of methylated CpG sites in the fragment of the sequencing data of the sample to be tested to the total number of CpG sites.

3. The method according to claim 1, characterized in that After analyzing the methylation status of the fragments using the fragment average methylation rate algorithm, spatial continuity algorithm, and control group baseline probability algorithm in the fragment analysis step, in the prediction step, based on at least one of the following information in the sequencing data of the sample to be tested: 1) the average methylation rate of all fragments in the target region, 2) the spatial continuity of all fragments in the target region, 3) the proportion of specific DNA fragments in the target region; predict whether the sample to be tested is derived from a diseased individual or a control group individual, or predict the probability that the sample to be tested is derived from a diseased individual.

4. The method according to claim 3, characterized in that The specific DNA includes tissue-specific DNA, cell-specific DNA or disease-specific DNA.

5. The method according to claim 3, characterized in that The specific DNA includes circulating cell-free tumor DNA.

6. The method according to claim 3, characterized in that The specific DNA fragment ratio refers to the ratio of specific DNA to all fragments in the sequencing data.

7. The method according to claim 1, characterized in that In the prediction step, the model is used to predict whether the sample to be tested is derived from a diseased individual or a control group individual, or to predict the probability that the sample to be tested is derived from a diseased individual.

8. The method according to claim 7, characterized in that The model includes a machine learning model or a probability distribution model.

9. The method according to claim 8, characterized in that The model is used to learn the training set feature weights of the diseased sample sequencing data and the control group sample sequencing data to identify the diseased samples.

10. The method according to claim 1, characterized in that The probability value of the sample to be tested coming from a diseased individual is determined based on the input features of the sample to be tested, and the probability value is compared with a preset probability threshold to determine whether the sample to be tested comes from a diseased individual.

11. The method according to claim 1, characterized in that In the analysis step, after using the fragment average methylation rate algorithm, spatial continuity algorithm, and control group baseline probability algorithm to analyze the methylation status of the fragment, in the prediction step, the average methylation rate of all fragments in the target area and the average value of at least one of the spatial continuity indicators are used as features, or the proportion of identified specific DNA fragments is used as a feature to construct a prediction model to predict whether the individual to whom the sample to be tested belongs has a disease, or the probability of the disease.

12. The method according to claim 1, characterized in that In the prediction step, the average methylation rate and spatial continuity of each retained abnormal fragment in the target region are calculated, and they are preprocessed into a two-dimensional matrix with the horizontal axis being the average methylation rate of each target region or the methylation level represented by spatial continuity and the vertical axis being each sample to construct a prediction model.

13. The method according to claim 1, characterized in that In the prediction step, if the target region is a disease hypomethylated region, only fragments whose spatial continuity is greater than a preset threshold or whose average methylation rate is less than a preset threshold are retained. The model is constructed using the proportion of the co-methylated fragment counts of the disease-specific hypomethylation in the methylated fragment counts of all quality control-qualified fragments. The model is preprocessed into a two-dimensional matrix with the horizontal axis being the specific DNA proportion of each target region and the vertical axis being each sample to construct a prediction model.

14. The method according to claim 13, characterized in that The specific DNA includes tissue-specific DNA, cell-specific DNA or disease-specific DNA.

15. The method according to claim 13, characterized in that The specific DNA includes circulating cell-free tumor DNA.

16. The method according to claim 13, characterized in that The specific DNA fragment ratio refers to the ratio of specific DNA to all fragments in the sequencing data.

17. The method according to claim 13, characterized in that The disease includes cancer or other diseases.

18. The method of claim 1, wherein: In the prediction step, the disease suffered by the patient includes cancer or other diseases.

19. The method according to claim 18, characterized in that In the methylation status acquisition step, the sample to be tested includes a body fluid sample or a tissue sample.

20. The method of claim 19, wherein: The body fluid sample includes plasma from blood, urine, saliva, bronchoalveolar lavage fluid or cerebrospinal fluid.

21. The method of claim 18, wherein: In the step of obtaining the methylation status, the methylation sequencing includes whole-genome methylation sequencing, degenerate methylation sequencing or whole-genome enzyme conversion sequencing.

22. The method according to claim 21, characterized in that In the step of obtaining the methylation status, the methylation sequencing is to convert the unmodified cytosine in the nucleic acid fragment into a base that does not bind to guanine or to convert the modified cytosine into a third modification by an enzyme.

23. The method of claim 22, wherein: In the step of obtaining the methylation status, the methylation sequencing is performed by treating the nucleic acid fragments through a non-enzymatic or enzymatic reaction.

24. The method of claim 22, wherein: The nucleic acid fragments include cfDNA or gDNA.

25. The method of claim 21, wherein: In the methylation status acquisition step, the methylation sequencing data is data aligned to a reference genome.

26. The method of claim 21, wherein: The prediction step includes obtaining multiple nucleic acid fragments from individuals in the control group, identifying the methylation status of each known CpG site based on the reference genome, and counting the corresponding methylated and unmethylated fragment counts to obtain the average methylation rate, and then estimating the corresponding parameters of the Bernoulli probability distribution or the beta probability distribution. The final cumulative multiplication result is the baseline probability of the control group.

27. The method of claim 26, wherein: In the prediction step, the nucleic acid fragments obtained from the control group individuals include cfDNA or gDNA.

28. The method of claim 26, wherein: In the methylation status acquisition step, the methylation status in the methylation sequencing data of the sample to be tested is acquired using the information extraction method of the co-methylated fragments.

29. The method of claim 26, wherein: The methylation status acquisition step includes obtaining the methylation information of the cfDNA fragments in the target region in the whole genome methylation sequencing data of the cfDNA of the sample to be tested.

30. The method of claim 26, wherein: In the methylation status acquisition step, the sample to be tested is derived from a human or an animal.

31. A device for predicting sample sources based on methylation differences, characterized in that: include: A methylation status acquisition module is used to obtain the methylation status in the methylation sequencing data of the sample to be tested; A fragment analysis module is used to analyze the methylation status of fragments in the sequencing data of the sample to be tested using a control group baseline probability algorithm, an average methylation rate algorithm, and a spatial continuity algorithm; A prediction module, used to predict whether the sample to be tested is derived from a diseased individual or a control group individual, or to predict the probability that the sample to be tested is derived from a diseased individual, based on the baseline data of the control group; In the prediction module, based on the relationship between the control group baseline probability value of the fragment in the sample to be tested and the preset threshold, it is predicted whether the fragment is an abnormal fragment different from the control group individual; In the prediction module, if the control group baseline probability value of the fragment in the sample to be tested is lower than a preset threshold, the fragment is predicted to be a fragment different from the control group individual; otherwise, the fragment is predicted to be a fragment derived from the control group individual; In the prediction module, the control group baseline data includes the average methylation rate corresponding to every 2, 3 or N adjacent CpG sites in the region, where N is a natural number greater than 3; In the prediction module, the control group baseline is a baseline established based on sample methylation sequencing data from individuals in the control group; In the fragment analysis module, the calculation formula for the baseline probability of the control group is as follows: （1） In formula (1), P(frag) is the control group baseline probability of the fragment, and P(S0) is the probability of the corresponding methylation state appearing at the first group of CpG sites in the fragment; P(S(n-1)n│S(n-1)) is the probability that the corresponding methylation state appears in the nth group of CpG sites in the fragment when the corresponding methylation state appears in the n-1th group of CpG sites in the fragment; In the fragment analysis module, the calculation formula for spatial continuity is as follows: （2） In formula (2), l is the total number of CpG sites covered by the fragment, N(i) From the start CpG site to the end CpG site of the fragment, it appears continuously i The number of sub-fragments with consistent specific methylation status at each CpG site, W i is the length weight of the sub-segment.

32. A device for predicting sample sources based on methylation differences, characterized in that: include: Memory, used to store programs; A processor, configured to implement the method according to any one of claims 1 to 30 by executing the program stored in the memory.

33. A computer-readable storage medium, characterized in that: The medium stores a program, and the program can be executed by a processor to implement the method according to any one of claims 1 to 30.

Images