Method for detection of samples with insufficient amount of fetal and circulating tumor DNA fragments for non-invasive genetic testing

The method addresses the challenge of detecting insufficient fetal or tumor DNA fragments in NIPT and liquid biopsy by using SNV counting and classification models to calculate DIFF distributions, enhancing the accuracy of DNA fraction estimation and clinical analysis.

WO2024242641A1PCT designated stage expired Publication Date: 2024-11-28MEDIREX GROUP ACADEMY N O
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
PCT/SK2023/000004
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-25
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

Current non-invasive prenatal testing (NIPT) and liquid biopsy methods face challenges in accurately detecting samples with insufficient amounts of fetal or circulating tumor DNA fragments, particularly those with fetal fractions below 5%, which can lead to incorrect evaluations and lack of reliable methods for estimation.

Method used

A method based on single nucleotide variant (SNV) counting, involving training and classification models to differentiate between sufficient and insufficient fetal or tumor DNA fractions through the calculation of DIFF distributions from fragment length and SNP data, using low-coverage whole genome sequencing data from cell-free DNA in plasma.

Benefits of technology

This approach enables accurate classification and estimation of fetal or tumor DNA fractions, improving the reliability of NIPT and liquid biopsy analyses by distinguishing between samples with sufficient and insufficient DNA amounts, thereby supporting clinical decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure SK2023000004_28112024_PF_FP_ABST
    Figure SK2023000004_28112024_PF_FP_ABST
Patent Text Reader

Abstract

The invention generally pertains to classification of samples with insufficient amount of fetal or circulating tumor DNA fraction based on single nucleotide polymorphisms. The invention is divided into two parts: training of classification model and evaluation of diagnoses samples. As a first step, blood from pregnant women was collected and prepared for DNA isolation and sequencing. BAM file was created and subsequently distributions for single nucleotide polymorphisms and coverage were built. Differences between these distributions determine the error and machine learning method was applied to the data to design a stable classifier. Samples were divided into training set and testing set and both sets are used to train and test a prediction model, respectively.
Need to check novelty before this filing date? Find Prior Art

Description

DescriptionTitle of Invention: Method for detection of samples with insufficient amount of fetal and circulating tumor DNA fragments for non- invasive genetic testingTechnical Field

[0001] The invention relates to the field of non-invasive genetic testing. The invention provides a method for detection of samples with insufficient amount of fetal and cir- culating tumor DNA fragments for non-invasive genetic tests based on analysis of single nucleotide variants detected via genomic sequencing.Background Art

[0002] Genome analysis

[0003] Genome is physically stored in a double helix DNA molecule, which consists of two strands, each carrying a sequence of nucleotides (A, T, G, C) called bases. A whole human genome is a sequence of roughly 3.2 billion DNA bases. The reference genome, artificial genome composed by scientists, is the most common sequence of bases in human DNA. Genome of every individual differs from the reference genome by around 0.5% of bases or less, owing to normal genetic variations and in specific cases (e.g., pregnancy and cancer) also when it comes to the difference between the so called maternal and placental (fetal) or normal and tumor genomes, respectively. These variations make each genotype unique and some of them can have significant impact on human health.

[0004] A standard procedure for obtaining individual variability is known. In the first step, collecting a biological sample is necessary (e.g., blood, saliva, etc.). The DNA molecule is then extracted and prepared for a process called sequencing. DNA se- quencing is a biochemical process for determining the precise order of nucleotide bases within the DNA molecule. When using the massive parallel sequencing technology, the molecule must be fragmented and placed on a sequencing platform. Here the fragments are read in parallel creating digital sequences of DNA bases called reads. These reads are randomly ordered with unknown direction and unknown DNA strand of origin. If genomic reads belong to an organism with a known reference genome, such as human, they can be sorted in the process called mapping.

[0005] Aim of the mapping is to reconstruct the original genomic sequence. Each read is aligned and thus mapped to the most probable region of origin on the reference genome. An aligned and mapped read is often simply called alignment. Set of aligned reads is de facto a digital copy of the DNA contained within the biological sample. Aligned reads reveal differences between a sequenced and the reference genome,called genomic variants. Whole set or even specific selection of genomic variants is unique to each individual hence a genome is the ultimate person identifier. Although the majority of these variants have no apparent effect on an individual, genome- wide association studies have linked some of the variants to diseases, appearance or even the behavior of an individual.

[0006] Variation of a single nucleotide at specific genomic locus between two genomes is called a single nucleotide variation (SNV). SNVs are the most abundant form of human genetic variation. If this variation is common in the general population it is more specifically called single nucleotide polymorphism (SNP). Depending on the presence of different variants at specific locus on a gene, the gene can have different alternative forms called alleles.

[0007] Human genome is organized into 22 pairs of homologous chromosomes and one pair of sex chromosomes. In each pair, one chromosome is derived from the mother and the other one from the father. The maternal and paternal chromosomes in a homologous pair have the same gene at the same locus, nevertheless alleles of this gene can differ between the chromosomes. If both alleles are identical, the organism is said to be ho- mozygous for that locus. If they differ, the organism is said to be heterozygous for that locus.

[0008] Prenatal testing

[0009] Prenatal testing for genomic defects of a fetus before birth is an integral component of current obstetric practice. The primary aim of the testing is screening for the presence of abnormal number of chromosomes. Testing is focused mainly on identifying an additional copy of a chromosome, the condition called trisomy (Allysc et al. 2015), represented as an excessive preportion of genomic material from the aberrant chromosome.

[0010] Discovery of fetal DNA fragments in plasma extracted from maternal blood, cell-free fetal DNA (cffDNA) (Y. M. D. Lo et al. 1997), opened up new options in the field of prenatal screening called non-invasive prenatal testing (NIPT) (BudiS et al. 2019; Y. M. Lo et al. 1997, BudiS et al. 2019). This fetal DNA can be detected from the fifth week of gestation and is collected with standard blood sampling from a mother.

[0011] Amniocentesis, traditionally used as a method of prenatal testing, requires amniotic fluid which needs to be extracted directly from the uterus of a mother. Therefore, am- niocentesis is an invasive procedure which puts mother and her fetus in risk of mis- carriage caused by needle injury, infection or stress.

[0012] The proportion of fetal fragments in analyzed DNA mixture is called fetal fraction. A sample with a trisomic fetus typically has an aberrant proportion of genomic material from tiie trisomic chromosome, compared to healthy samples. A trisomic sample with a very low fetal fraction can, however, be incorrectly evaluated as healthy, since theaberrant chromosome may cause only a weak deviation from normal values that can be presumed to represent a normal measurement error (Ashoor et al. 2013), which is par- ticularly problematic for the detection of sub-chromosomal aberrations. Fetal fractions below 5% are set to be uninformative in NIPT. Since there are no current trustworthy methods for estimating samples with low fetal fraction (< 5%), the reliable method to identify samples with an insufficient fetal fraction is essential.

[0013] According to the current knowledge, fetal fragments cannot be unambiguously dis- tinguished from the maternal ones. For these reasons NIPT still does not achieve accuracy of invasive methods like amniocentesis, and further improvements arc necessary for their full replacement. In contrast to established screening methods (Alfirevic, Mujezinovic, and Sundberg 2003), sampling of genetic material from the mother’s circulation does not pose any direct risk for the fetus (Bianchi et al. 2014).

[0014] The reliability of non-invasive prenatal testing (NIPT) is highly dependent on the accurate estimation of fetal fraction. Several methods have been proposed up to date, utilizing different attributes of analysed genomic material, for example length and genomic location of sequenced DNA fragments ((Straver et al. 2016). In particular, pregnant samples with low fetal fraction are highly demanding to distinguish from non-pregnant samples to identify potential exchange of samples or contamination. Here we present a method for classification of samples based on SNP counts per fragment which is, according to our knowledge, not published yet.

[0015] Count based methods which calculate disproportion of the number of reads mapped to chromosomes differ in mother and fetus genotypes. Although they are quite reliable, they can be used only on samples with male (XX vs. XY genotype) (Fan et al. 2008) or trisomic (2 copies vs. 3 copies of an aberrant chromosome) fetuses (Chiu et al. 2011; Dims et al. 2018). In pregnancy with a healthy female fetus, pregnancy has to be de- termined by alternative methods.

[0016] As fetal DNA has generally shorter fragments than maternal DNA, authors of fragment length (FL) model presume that maternal plasma samples with a higher fetal DNA fraction would have a higher proportion of short DNA fragments. FL model is then based on the size analysis of short and long DNA fragments in the maternal plasma with the assumption that ratio of short DNA fragments and long DNA fragments is correlated to proportion of fetal DNA fragments in maternal plasma. Dataset of measured values (read lengths distribution of each mother) is then divided into two groups: a training group and a validation group and together they are used to create a linear regression representing the resulting fetal fraction from read lengths dis- tribution ((Yu et al. 2014)(Y. M. Lo et al. 1997; Gazdarica et al. 2019)(Yu et al. 2014).

[0017] Alternatively, fetal fraction may be estimated from the genomic location patterns that slightly differ between maternal and fetal fragments due to differences of their nu-cleosome positioning (SANEF ALCON method) (Straver et al. 2016) and euchromatic DNA structure (SeqFF method) (Kim et al. 2015). Due to high accuracy and no ad- ditional laboratory costs, the SeqFF method is a preferred method for samples with female fetus (van Seek et al. 2017). The method takes as input a vector of Loess- corrected fragment counts partitioned into bins 50,000 bases long. The fetal fraction is then determined using standard multivariate regression models. Estimation of these weights requires a huge amount of training samples that are scarcely available for small laboratories, and so the method is applicable only for established tests with a large cohort of pre-analyzed samples.

[0018] The S ANEFALCON method uses more detailed information of fragment origin based on molecular mechanisms of fragment degradation that differ between maternal and fetal fragments. At first, fragments from all training samples are aggregated together, to localize local coverage peaks. Authors assumed that these peaks indicate positions of chromosomal packing elements, nucleosomes, that stabilize DNA fragments. Fragment length profiles differ slightly between fetal and maternal DNA due to differences in DNA degradation. Published results supported this hypothesis, however the accuracy of the method is not sufficient for reliable fetal fraction prediction or sample typing.

[0019] Patent AU2011218382B2 entitled "Methods for detecting fetal nucleic acids and di- agnosing fetal abnormalities" by Efcavitcb et al. claims detecting the presence or absence of fetal nucleic add using a technique selected from sparse allele calling, targeted gene sequencing, identification of Y chromosomal material, enumeration, copy number analysis, and inversion analysis.

[0020] Patent US9493828B2 entitled "Methods for determining fraction of fetal nucleic acids in maternal samples" by Rava et al. claim gender-independent fetal fraction es- timation method. It comprises using a predetermined set of SNP positions, counting reads mapped to these positions and identifying at least 10 informative SNPs by the allelic difference between the reads for each SNP. Fetal fraction is calculated from the total number of mapped reads to a first allele and the total number of reads mapped to a second allele at each of the said informative SNPs.

[0021] Lo et al. in their patent US10208348B2 entitled "Determining percentage of fetal DNA in matonal sample" claim fetal fraction estimation method. Specifically, they claim a method for determining whether genomic imbalance such as chromosomal aneuploidy exists within a biological sample based on a fetal fraction threshold. The fetal fraction is computed from the same or different data used to determine the cut-off value. The threshold value can be computed from the formula: (3xF+2x(l-F)) / (2xF+2x(l-F)), where F is the fetal fraction. The method uses sites of genetic polymorphism where the mother is homozygous and the fetus is heterozygous.More specifically, these are sites where the fetus has one identical and one different allele when compared to a maternal pair of identical alleles, giving two different alleles in total. The fetal fraction is computed by comparison of the number of reads supporting the first allele and the number of reads supporting the second allele for a specific locus.

[0022] The method most similar to the proposed invention is FetalQuantSD (Jiang et at 2016) a variation of a previous method FetalQuant by the same authors (Jiang et ah 2012). These methods for fetal fraction estimation are based on the differentiation of variants detected in maternal and child DNA. Although several methods for fetal fraction estimation have been proposed, there are no current methods that detect samples with insufficient fetal fraction.

[0023] The FetalQuant method estimates fetal fraction and as the input it needs deeply sequenced mother cells (purely mother DNA) and deeply sequenced cfDNA of pregnant women. The deep sequencing requirement is needed to be able to identify variants both for mother and for mixture of fetus-mother DNA. These variants are then stored in compact VCF formats and further analyzed. Essentially, the variants that arc present in the mother-fetus mixture and not present in mother are categorized as fetus variants inherited from father and based on their frequency the fetal fraction is estimated. This fetal fraction is then cleaned from effects of sequential errors and other systematic bias.

[0024] The FetalQuantSD method improves the previous method, since it no longer relies on the deeply sequenced cfDNA of pregnant women, but only shallow sequencing is needed. However, deep sequencing of purely mother variants is still needed and the accuracy of the fetal fraction estimation is highly dependent on the number of SNP loci identified in the mother and on the number of reads of the shallow cfDNA sequencing.

[0025] The FetalQuantSD computes the fetal fraction as follows: first, it compiles a list of homozygous SNPs in the mother from the first data set. Then, it searches for non- matemal alleles in reads from the second data set These non-matemal alleles would potentially represent paternally inherited fetal alleles. However, a small proportion of these non-matemal alleles could be caused by sequencing errors in maternal plasma and / or genotyping errors in maternal genomic DNA. The authors assume linear error rates across different cases and correct it by a linear regression on a training set composed of 23 samples. The actual DNA fraction was deduced by comparing the aligned sequence reads to the sites where the maternal genotypes were homozygous (AA) and the fetal genotypes were heterozygous (AB) using the formula F = 2p / (p+q), where p is the number of sequenced reads carrying fetal-specific alleles (i.e., allele B) and q is the number of sequenced reads carrying alleles shared by the mother and the fetus (i.e., allele A).

[0026] This method uses only homozygous SNP in the mother and needs a significant number of mother SNPs and sequencing reads of the cfDNA to achieve satisfactory results.

[0027] Liquid biopsy in cancer

[0028] DNA fragments of fetal origin present in the maternal circulation have been shown to be similar to DNA fragments of tumor origin in the circulation of individuals with cancer. This phenomenon has been definitively confirmed and the term liquid biopsy has been established in the context of analysis focused on non-germline representing DNA fragments originated in cancer cells (also relevant for other types of nucleic acids, e.g., RNA). The concept of liquid biopsy in cancer and its worldwide research is quite closely linked to the discoveries in NIPT and the use of next-generation se- quencing, with key findings published by the same research group and used as a reference to this day (Chan, KC Allen, et aL 2013). The same principle used in the analysis of cell-free fetal DNA (cffDNA) from plasma from pregnant women could be practically directly applied to the detection and determination of circulating tumor DNA (ctDNA), as both have similar biological properties (e.g., a specific size of cir- culating DNA fragments, the proportion of circulating fetal / tumor DNA corresponding to the size of placenta / amount of tumor cells). The main difference concerns the origin of the DNA fragments, whereas in NIPT they are represented by DNA fragments of maternal (maternal body tissues) and fetal (placental) origin, in liquid biopsy the DNA fragments of interest are of normal and tumor tissue origin. Similarly, they differ in the presence of genomic alterations representing all types of variants (CNVs, SNVs, indels), which can be exploited in both qualitative and quantitative analyses using the same methods and analytical tools. The present invention is therefore applicable to both fields, replacing some of the terms used (e.g., paternal alleles in NIPT for somatic mutations in cancer-related liquid biopsies). From a practical point of view and in order to avoid duplication of information, the proposed invention is more oriented towards NIPT but is fully applicable to cancer-related liquid biopsy analyses.Summary of Invention

[0029] The presented invention provides an alternative method for determination of sample typing for non-invasive genetic testing of pregnant women or cancer suspected in- dividuals, specifically detection of samples with insufficient amount of fetal or cir- • dilating tumor DNA fraction (further also briefly “insufficient samples”). This method is based on the SNP counting approach, thus it can be potentially combined with other methods based on different principles to get more reliable results.

[0030] The method for detection of insufficient amount of fragments of fetal (placental) / tumor origin in samples of peripheral plasma of pregnant women / cancer suspected in-dividuals within whole genome analysis based non-invasive genetic tests, according to present invention, is comprising of the steps of a) training of classification model, comprising the steps of, i) collection of plasma samples with sufficient and insufficient amount of fetal / tumor fragments, ii) separation of plasma fraction from whole blood sample, iii) isolation of cell-free DNA from plasma samples, iv) performing a random sequencing of the cell-free DNA molecules contained in the plasma, v) determination of fragment length distribution of all DNA fragments, vi) identification of sequenced DNA fragments genomic coordinates and determination of present genomic variants, vii) determination of fragment length distribution of individual DNA fragments with genomic variant, viii) calculating the DIFF distribution as a difference between these two distributions, ix) training of classification model on the DIFF distributions, where predicted variable is sufficient or insufficient amount of fetal / tumor fragments; and b) evaluation of a diagnosed sample, comprising the steps of i) calculating DIFF dis- tribution of the sample using the same steps i)-vii) as for the said training of classi- fication model, ii) using pre-trained classification model to predict sufficient or in- sufficient amount of fetal / tumor fragments in the diagnosed sample.

[0031] The method according to the invention can comprise the calculation of the fetal / cir- culating tumor DNA fraction according to the DIFF distribution.

[0032] According to an aspect of the present invention a computer system is provided, comprising at least one computing device and / or server which comprises one or more processors and one or more modules configured to execute the method according to the present invention.

[0033] According to an aspect of the present invention a computer program product comprising computer-readable instructions is provided. The computer program product, when loaded into and executed on a computer system, causes the system to execute operations according to the method according to the present invention. Brief Description of Drawings

[0034] [Fig.l] Shows the difference in the median distribution of DIFF for certain length.

[0035] [Fig.2] Representation of test samples using the Nearest neighbor classification method showed a clear classification. Circles and dots represent prediction and ground truth, respectively. Samples with sufficient FF are represented with 1 and 0 for samples with insufficient FF.Description of the invention

[0036] The following description will explain the main features of the method of the present invention; however, it does not imply that the invention must include all features and aspects described herein. The skilled person will get a full understanding of the present invention from the following description together with the examples, where somespecific features and aspects will be explained in more detail.

[0037] The technical and scientific terms used herein have the same meaning as commonly understood by the persons skilled in the art of medicine, molecular genetics, prenatal non-invasive diagnostics, liquid biopsy, molecular biology and bioinformatics. Some specific terms are explained in the “Definitions” section below.

[0038] The method of the invention comprises steps on the level of biology (molecular biology) and bioinformatics:1) Biological steps a) Obtaining samples of blood b) Preparation of DNA sample and DNA library c) Sequencing2) Bioinformatical steps a) Mapping b) Processing of the prepared data to detect the insufficient samples

[0039] Only these two inputs are required - low coverage WGS of cell-free DNA from blood and reference genome. The coverage of the reads is typically in the range between 0.2 - lx. The reads should be preferably in a FASTQ file format.

[0040] This invention does not rely on an exact method for obtaining sequencing and mapping data, therefore this gathering process is not described here in detail. However, people skilled in the field of molecular biology know how to get these data using one of the available technologies (Illumina sequencing, lonTorrent sequencer, MGI sequencer, ...). In all our tests and examples, the data comes from NextSeq5OO system sequencer (Illumina, Inc., San Diego, CA, USA).

[0041] Definitions

[0042] Variant The term variant herein refers to a difference between a genome and the reference genome. It is defined by start position (on the reference), reference allele and alternative allele. In other words, it is a replacement of a reference allele by an al- ternative allele at a specific genomic position.

[0043] Allele: Depending on the presence of one or more variants, the same gene can have different forms, called alleles. Here, the term allele refers to a particular sequence at the position of a variant. Reference allele is the sequence present in the reference genome. Alternative allele is a different sequence in place of reference allele. Variant is always described by one reference allele and at least one alternative allele.

[0044] Zygosity: Since a human has two sets of homologous chromosomes, it is a diploid organism, which implies it has two copies of each allele, except for the sex chromosomes. If both alleles of a diploid organism are identical, the organism is said to be homozygous for that position. If they differ, the organism is said to be het- erozygous for that position.

[0045] Genotype: Set of all alleles in the genome.

[0046] Genome: The complete set of DNA sequences within an organism.

[0047] Massively parallel sequencing (MPS) or next generation sequencing (NGS):Techniques, that are well known to the people skilled in the art of sequencing of DNA. The most widely known sequencing technologies in the field are Illumina or lonTorrent Protocols for whole genome sequencing are known to the skilled person and can be found in the examples below.

[0048] (Sequencing) reads: A read is an inferred sequence of base pairs (or base pair proba- bilities) corresponding to all or part of a single DNA fragment In other words, they are small continuous parts of an individual's DNA. The read should be long enough to serve as a sequence tag, so it can be unambiguously mapped or assigned to a precise location to a reference genome - at least 30-35bp. Usually a small degree of mismatch (Ibp) can encompass natural variation in individual genomes.

[0049] Deep vs shallow sequencing: Deep sequencing means to sequence with a very high coverage of the sequenced part (usually 30x and more for variant analysis). With WGS, this means sequencing of roughly hundreds of millions to billions of reads. Shallow sequencing commonly used in clinical practice especially in modem non- invasive prenatal testing on the other hand uses WGS of only 0.2x - 0.5x coverage so it is in the order of millions to few tens of millions of reads.

[0050] Mapping: Alignment of the sequence information from NGS (i. e. DNA fragment the genomic position of which is unknown) with a matching part in reference to the human genome. This can be done in several ways; we used the method of Liu et al 2014. As used herein, the term human reference genome or reference genome refers to Grch38 sequence. Reads that do not map uniquely (map to several positions) are usually excluded from the analysis. The alignment is usually done by computer al- gorithms well known to the persons skilled in the art of molecular biology and bioin- formatics.

[0051] SAM / BAM file: File that contains aligned sequencing reads in a text format (SAM) or a compressed binary format (BAM). For every read it contains its mapped position on the reference genome (if the mapping for that read was successful), the mapping quality, sequencing quality (if provided), the location of the paired read (in case of pair-end sequencing), and various other information. It is a standard for storing aligned reads. Each SAM / BAM file is dependent on the reference genome used - this in- formation is stored in the header of the SAM / BAM file.

[0052] FASTQ file: File containing all reads from a sequencer, together with its sequencing quality. This is the standard file format to store this data and it is usually compressed to save disk space. All the modem mapping software accept this format as input

[0053] WGS: Whole genome sequencing (WGS) strives to read all genomic informationwithin a DNA sample. It is associated with high cost and high data yield.

[0054] For non-invasive genetic testing, blood from pregnant women or cancer suspected in- dividuals is collected and prepared for DNA isolation and sequencing. After se- quencing, sequencing reads are aligned to the human reference genome (for example with bowtie2 mapping tool (Langmead and Salzberg 2012).

[0055] Then, individual BAM files of pregnant / nonpregnant / cancer suspected / healthy samples are created using specific bioinformatic tools. Each read is checked for SNP positions and summed with other reads of the same length containing SNP position. This is similarly done for coverage. Difference between relative coverage and relative count of SNP per fragment of certain length is calculated to determine how far from each other these distributions are. Statistical or machine learning methods applied to the data leads to determination of criterion for detection of samples with insufficient fetal / circulating tumor DNA fraction (FF / ctDNAF).

[0056] Sample Preparation and Sequencing

[0057] Blood from pregnant women or cancer suspected individuals is collected. Blood plasma is separated after collection and prepared for DNA isolation. Standard fragment libraries for massively parallel sequencing are prepared (for example using Illumina NextSeq 500 / 550 High Output Kit v2 (San Diego, CA, USA)

[0058] Mapping and Read Count Correction

[0059] Sequencing reads are aligned to the human reference genome (Grch38). Sub- sequently, FASTQ files are mapped to human genome reference (preferably version GRCh38.pl0), using Bowtie2 (v2.1.0) (Langmead and Salzberg, 2012) resulting in one Sequence Alignment Map (SAM) file for each sample. SAM files are converted to Binary Alignment Map (BAM) format, sorted and indexed using Samtools view, merge and index utilities (v0.1.19) (Li et at, 2009).

[0060] Sample classification for detection of insufficient samples

[0061] BAM file consisting of pregnant / nonpregnant / cancer suspected / healthy samples is processed using in-house scripts to count SNPs. First, reads of length 50-200 bp are filtered out. Tensors for SNPs of dimensions number of chromosomes x number of samples x fragment length filled with zeroes are created. Then each read is checked and counted SNP positions for the read and summed it with other SNP counts per- read of certain length. Similarly, tensors of coverage for reads of a certain length are built up. Read lengths are summed to determine coverage for fragments of certain length.

[0062] Then, following parameters are deterrnined / calculatedRelative counts of SNPs per fragment of certain length - SNPi, i = 50, 51. 200Relative coverage per fragment of certain length -COVi, i = 50, 51, ..., 200 DIFFi = absolute value (COVi - SNPi)

[0063] In the FL model, the FF / ctDNAF is estimated from the specific ratio of the relativerepresentation of short and long DNA fragments. This estimation is possible because the authors showed that there are significant differences between the pregnant women in different gestational ages and non-pregnant samples, for different segments of the length intervals. Recently published original studies showed the same characteristics of ctDNA with shorter fragments representation of tumor and longer of normal cells originated DNA fragments (Markus, Havell, et al. 2022). Further refinements of this method compare the whole distribution of length profiles, not just a specific ratio.[00641 In this invention, instead of representing fragment lengths, we compare the repre- sentation of differences in coverage and number of SNPs for fragment specific lengths.[00651 As also reported in Gazdarica et at 2019, FF has an effect on the length of DNA fragments circulating in the blood. It has been shown that fetal fragments tend to be shorter than maternal fragments. This means that the distributions, or their repre- sentation, for individual lengths vary depending on whether fetal fragments are observed in the mixture and also how many of them are present Statistical and machine learning methods are usually used to estimate the parameters of the observed model. Here, the amount of the FF was estimated. Analyses are usually unsuccessful, also called uninformative, if the value of the estimated FF is below the threshold of 5% (0.05), according to more stringent criteria and with the use of a larger number of se- quencing reads, this threshold can be reduced to 4% (0.04). The methods used are mainly regressors and mathematical methods that can estimate not only whether fetal fragments are present in the blood, but also how many of than are present. Commonly used methods:1. Linear Regression2. Logistic Regression3. Ridge Regression4. Lasso Regression5. Polynomial Regression6. Bayesian Linear Regression7. XGBoost8. Support vector machine9. Nearest neighbor10. Randan Forest

[0066] This is followed by testing the accuracy, or error, of the FF estimation, as the re- liability of the NIPT depends on this. Accurate FF estimation is one of the key at- tributes and must not be underestimated. It is not rare that results at the FF cutoff are often questionable or uninformative, precisely because of concerns about the correct estimation of the FF. These results need to be supported.

[0067] In the case of this invention, a classification problem is involved. The aim is todetermine whether the sample has sufficient FF to make it worthwhile to be analysed further, or to support the questionable results of standard NIPTs for which the FF is around the borderline, and thus to give clinicians additional information to use in their decision making. Similarly, the same could be applied to liquid biopsy analysis focused on ctDNA with ctDNAF as a limiting parameter.

[0068] One can easily see the difference in the median of DIFF distributions between samples with sufficient and insuflident FF, respectively, [Hg.l], Significant dif- ferences in certain fragment lengths are observed, which are related to the amount of FF.

[0069] Statistical or machine learning methods suitable for classification are applied to DIFFis data of certain length to finally classify samples into samples with sufficient and insufficient amount of fetal fraction.Examples

[0070] Preparation of samples (NIPT)

[0071] Sample Preparation and Sequendng

[0072] Blood from pregnant women was collected. Blood plasma was separated after collection and prepared for DNA isolation. DNA isolation is performed according to TruSeq Nano Protocol (Illumina). Standard fragment libraries for massively parallel sequencing were prepared. Library preparation was performed according to TruSeq Nano Protocol (Illumina). Illumina NextSeq 500 / 550 High Output Kit v2 (San Diego, CA, USA) (75 cycles) was used for massively parallel sequencing of prepared libraries using pair-end sequendng with read length of 2x35bp on an Dlummina NextSeq 500 platform (Available online: https: / / www.illumina.com / ). Samples were sequenced on NextSeq 500 where each read has length 35bp and is read in both directions and output size is approximately 300-700 MB per sample generated in FASTQ format.

[0073] Mapping and Read Count Correction

[0074] Sequencing reads were aligned to the human reference genome (hgl9). Sub- sequently, FASTQ files were mapped to human genome reference, version GRCh38.pl0, using Bowtie2 (v2.1.0) (Langmead and Salzberg, 2012) resulting in one Sequence Alignment Map (SAM) file for each sample. SAM files were converted to Binary Alignment Map (BAM) format, sorted and indexed using Samtools view, merge and index utilities (v0.1.19) (Li et al., 2009).

[0075] Device

[0076] The content of the patent was created for the purpose of more accurate evaluation of NIPT and cancer related liquid biopsy analyses. This content has been uploaded to a device that serves this purpose. This device first maps reads from sequenced sample to a human reference genome with Bowtie2 tool (Langmead and Salzberg 2012) and socreating file (BAM) represents aligned sequences. BAM file is further processed to obtain a number of SNP positions for each read.

[0077] In the next step BAM file consisting of a pregnant or nonpregnant sample / healthy or cancer samples using in-house scripts is processed to count SNPs and fragment length distributions. Reads of length 50-200 bp are filtered out to speed up the cal- culations.

[0078] Tensors for SNPs are constructed by these dimensions:Number of chromosomes x Number of samples x Fragment length filled with zeroes are created and consequently the device checked each read and counted SNP positions for the read and finally summed it with other SNP counts per read of certain length. Similarly, tensors of coverage for reads of a certain length were built up. The device summed read lengths to determine coverage for fragments of certain length.

[0079] Relative counts of SNPs per fragment of certain length and relative coverage per fragment of certain length are derived and absolute value of the differences between relative counts of SNPs and coverage are calculated with the device asDIFFi = absolute value(COVi - SNPi) for i = 50, 51 200.

[0080] Here the Nearest neighbor classifier (NNC) applied to DIFFis of certain length to classify the samples. NNC is a machine learning method that aims at labeling previously unseen query objects while distinguishing two or more destination classes.

[0081] The output of this program is classification whether the sample is sufficient or in- sufficient for NIPT, in other words, detection of the sample with insufficient fetal fraction. Again, the same could be applied to liquid biopsy analysis focused on ctDNA with ctDNAF as a limiting parameter.

[0082] Veriflcation / vaHdation of the classification model

[0083] Training and testing data set

[0084] We collected 63 samples from which 49 and 14 samples were pregnant and non- pregnant women, respectively. 20 of the pregnant samples had critical amounts of FF (FF between 3% and 6%). These samples were sequenced in a row in the same run to avoid any possible sequencing bias. For each such experiment, the whole dataset (63 samples) was divided into 80% for training and 20% for testing. In other words, the presented method in this paper was evaluated on 13 testing and 50 training samples, re- spectively. We used a threshold value of 4% in the evaluation. FF with a value less than 4% is considered insufficient For FF greater than or equal to 4%, the sample has sufficient FF and is ready for further analysis. We applied the NNC to classify these testing samples resulting in an unambiguous classification, [Fig.2],

[0085] The aim is to determine whether the sample has sufficient FF to make it worthwhile to be analysed further, or to support the questionable results of standard NIPTs forwhich the FF is around the borderline.

[0086] We have shown that a method based on displaying differences in coverage and number of SNPs for fragments of a specific length is undoubtedly a useful tool to gain new information from an uninformative sample due to a low FF, with the aim of fa- cilitating further decision making by clinicians.Citation List

[0087] Citation List follows:Patent Literature

[0088] PTL 1: AU2011218382B2

[0089] PTL 2: US9493828B2

[0090] PTL 3: US10208348B2Non Patent Literature

[0091] NPL 1 : Alfirevic, Zarko, Faris Mujezinovic, and Karin Sundberg. 2003. “Amniocentesis and Chorionic Villus Sampling for Prenatal Diagnosis.” Cochrane Database of Systematic Reviews, https: / / doi.org / 10.1002 / 14651858.cd003252.

[0092] NPL 2: Allyse, Megan, Mollie A. Minear, Elisa Berson, Shilpa Sridhar, Margaret Rote, Anthony Hung, and Subhashini Chandrasekharan. 2015. “Non-Invasive Prenatal Testing: A Review of International Implementation and Challenges International Journal of Women's Health 7 (January): 113-26.

[0093] NPL 3: Ashocr, G., A. Syngelaki, L. C. Y. Poon, J. C. Rezende, and K. H. Nicolaides. 2013. “Fetal Fraction in Maternal Plasma Cell-Free DNA at 11-13 Weeks' Gestation: Relation to Maternal and Fetal Characteristics." Ultrasound in Obstetrics & Gy- necology. https: / / doi.org / 10.1002 / uog.12331.

[0094] NPL 4: Beck, Daphne M. van, Roy Straver, Marian M. Weiss, Elies M. J. Boon, Karin Huijsdens-van Amsterdam, Cees B. M. Oudejans, Marcel J. T. Reinders, and Brik A. Sistermans. 2017. “Comparing Methods for Fetal Fraction Determination and Quality Control of NIPT Samples.” Prenatal Diagnosis, https: / / doi.org / 10.1002 / pd.5079.

[0095] NPL 5: Bianchi, Diana W., R. Lamar Parker, Jeffrey Wentworth, Rajeevi Madankumar, Craig Saffer, Anita F. Das, Joseph A. Craig, et al. 2014. “DNA Se- quencing Versus Standard Prenatal Aneuploidy Screening." Obstetrical & Gyne- cological Survey. https: / / doi.org / 10.1097 / 0Logx.0000451481.18231. lb.

[0096] NPL 6: BudiS, Jaroslav, Marcel Kucharik, FrantiSek 6uriS, Juraj Gazdarica, Michaels Zrubcovd, Andrej Ficek, TomfiS Szemes, Brofia Brejovd, and Jan Radvanszky. 2019. “Dante: Genotyping of Known Complex and Expanded Short Tandem Repeats.'' Bioinformatics 35 (8): 1310-17.

[0097] NPL 7: Chiu, R. W. K., R. Akolekar, Y. W. L. Zheng, T. Y. Leung, H. Sun, K. C. A. Chan, F. M. F. Lun, et al. 2011. “Non-Invasive Prenatal Assessment of Trisomy 21 byMultiplexed Maternal Plasma DNA Sequencing: Large Scale Validity Study." BMJ. https: / / doi.org / 10.1136 / bmj .c7401.

[0098] NPL 8: Chan, K. A., Jiang, P., Zheng, Y. W., Liao, G. J„ Sun, H., Wong, J & Lo, Y. M. D. (2013). Cancer genome scanning in plasma: detection of tumor-associated copy number aberrations, single-nucleotide variants, and tumoral heterogeneity by massively parallel sequencing. Clinical chemistry, 59(1), 211-224.

[0099] NPL 9: Doris, Frantisek, Juraj Gazdarica, Iveta Gazdaricova, Lucia Strieskova, Jaroslav Budis, Jan Tuma, and Tomas Szemes. 2018. “Mean and Variance of Ratios of Proportions from Categories of a Multinomial Distribution." Journal of Statistical Dis- tributions and Applications, https: / / doi.org / 10.1186 / s40488-018-0083-x.

[0100] NPL 10: Fan, H. Christina, Yair J. Blumenfeld, Usha Chitkara, Louanne Hudgins, and Stephen R. Quake. 2008. “Noninvasive Diagnosis of Fetal Aneuploidy by Shotgun Se- quencing DNA from Maternal Blood.” Proceedings of the National Academy of Sciences of the United States of America 105 (42): 16266-71.

[0101] NPL 11 : Gazdarica, Juraj, Rastislav Hekel, Jaroslav Budis, Marcel Kucharik, Frantisek Duris, Jan Radvanszky, Jan Tuma, and Tomas Szemes. 2019. “Combination of Fetal Fraction Estimators Based on Fragment Lengths and Fragment Counts in Non-Invasive Prenatal Testing." International Journal of Molecular Sciences 20 (16). httpsV / doi.org / 10.3390Ajms20163959.

[0102] NPL 12: Jiang, Peiyong, K. C. Allen Chan, Gary J. W. Liao, Yama W. L. Zheng, Tak Y. Leung, Rossa W. K. Chiu, Yuk Ming Dennis Lo, and Hao Sun. 2012. “FetalQuant: Deducing Fractional Fetal DNA Concentration from Massively Parallel Sequencing of DNA in Maternal Plasma." Bioinformatics 28 (22): 2883-90.

[0103] NPL 13: Jiang, Peiyong, Xianlu Peng, Xiaoxi Su, Kun Sun, Stephanie C. Y. Yu, Weng In Chu, Tak Y. Leung, et al. 2016. “FetalQuantSD: Accurate Quantification of Fetal DNA Fraction by Shallow-Depth Sequencing of Maternal Plasma DNA.” Npj Genomic Medicine, https: / / doi.org / 10.1038 / npjgenmed.2016.13.

[0104] NPL 14: Kim, Sung K., Gregory Hannum, Jennifer Geis, John Tynan, Grant Hogg, Chen Zhao, Taylor J. Jensen, et al. 2015. “Determination of Fetal DNA Fraction from the Plasma of Pregnant Women Using Sequence Read Counts.” Prenatal Diagnosis 35 (8): 810-15.

[0105] NPL 15: Langmead, Ben, and Steven L. Salzberg. 2012. “Fast Gapped-Read Alignment with Bowtie 2.” Nature Methods, https: / / doi.org / 10.1038 / nmeth.1923.

[0106] NPL 16: Lo, Y. M., N. Corbetta, P. F. Chamberlain, V. Rai, I. L. Sargent, C. W. Redman, and J. S. Wainscoat 1997. “Presence of Fetal DNA in Maternal Plasma and Serum." The Lancet 350 (9076): 485-87.

[0107] NPL 17: Lo, Y. M. Dennis, Y. M. Dennis Lo, Noemi Corbetta, Paul F. Chamberlain, Vik Rai, Ian L. Sargent, Christopher W. G. Redman, and James S. Wainscoat 1997.“Presence of Fetal DNA in Maternal Plasma and Serum.’’ The Lancet https: / / doi.org / 10.1016 / s0140-6736(97)02174-0.

[0108] NPL 18: Markus, H„ Chandrananda, D., Moore, E., Mouliere, F., Morris, J., Brenton, J. D., ... & Rosenfeld, N. (2022). Refined characterization of circulating tumor DNA through biological feature integration. Scientific Reports, 12(1), 1928.

[0109] NPL 19: Straver, Roy, Cees B. M. Oudejans, Erik A. Sistermans, and Marcel J. T. Reinders. 2016. “Calculating the Fetal Fraction for Noninvasive Prenatal Testing Based on Genome-Wide Nucleosome Profiles.” Prenatal Diagnosis. https: / / doi.org / 10.1002 / pd,4816.

[0110] NPL 20: Yu, Stephanie C. Y., K. C. Allen Chan, Yama W. L. Zheng, Peiyong Jiang, Gary J. W. Liao, Hao Sun, Ranjit Akolekar, et al. 2014. “Size-Based Molecular Di- agnostics Using Plasma DNA for Noninvasive Prenatal Testing." Proceedings of the National Academy of Sciences of the United States of America 111 (23): 8583-88.

Claims

Claims

1. A method for detection of insufficient amount of fragments of fetal (placental) / tumor origin in samples of peripheral plasma of pregnant women / cancer suspected individuals within whole genome analysis based non-invasive genetic tests characterized in that the method is comprising the steps of: a) Training of classification model, comprising the steps of i) Collection of plasma samples with sufficient and insufficient amount of fetal / tumor fragments ii) Separation of plasma fraction from whole blood sample iii) Isolation of cell-free DNA from plasma samples iv) Performing a random sequencing of the cell-free DNA molecules contained in the plasma v) Determination of fragment length distribution of all DNA fragments vi) Identification of sequenced DNA fragments genomic coordinates and determination of present genomic variants vii) Determination of fragment length distribution of individual DNA fragments with genomic variant viii) Calculating the DIFF distribution as a difference between these two distributions ix ^Training of classification model on the DIFF distributions, where predicted variable is sufficient or insufficient amount of fetal / tumor fragments; and b) Evaluation of a diagnosed sample, comprising the steps of i) Calculating DIFF distribution of the sample using the same steps i)-vii) as for the training of classification model ii) Using pre-trained classification model to predict sufficient or in- sufficient amount of fetal / tumor fragments in the diagnosed sample

2. The method according to claim 1 comprising the calculation of the fetal or circulating tumor DNA fraction according to the DIFF distribution.

3. A computer system comprising at least one computing device and / or server, which comprises one or more processors and one or more modules configured to execute the method according to any one of the claims 1 and 2.

4. A computer program product comprising computer-readable in- structions which, when loaded into and executed on a computer system, causes the system to execute operations according to the methodaccording to any one of the claims 1 or 2.

Citation Information

Patent Citations

  • Determining percentage of fetal DNA in maternal sample

    US10208348B2

  • Methods for determining fraction of fetal nucleic acids in maternal samples

    US9493828B2

  • Limit of detection based quality control metric

    US20210366569A1

  • AU2011218382B2