Methods for accurate computational decomposition of DNA mixtures from unknown genotypic contributors

The challenge of quantitative analysis of contributors in nucleic acid mixture samples was solved through probabilistic mixture modeling and Bayesian inference techniques, achieving more accurate deconvolution and quantification, especially in the case of unknown genotypes.

CN110770839BActive Publication Date: 2024-12-06ILLUMINA INC
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN201880041562.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2017-06-20
Filing Date
2018-06-19
Publication Date
2024-12-06
Estimated Expiration
2038-06-19

AI Technical Summary

Technical Problem

In the research and clinical settings, there are challenges in quantifying contributors in sequencing data from nucleic acid mixtures from tightly related genomes, especially when the original genome is unknown, existing methods are difficult to accurately deconvolution and quantify.

Method used

Using probability mixture model and Bayesian inference technology, the nucleic acid fraction in nucleic acid samples is quantified through computer systems, the errors in the nucleic acid sequence reading are modeled using probability distribution, and the nucleic acid of the contributor is determined by maximizing the likelihood function. Fraction.

Benefits of technology

Improves the analytical sensitivity and specificity of nucleic acid mixture samples, enables accurate deconvolution and quantification, and even provides accurate analysis results when the contributor genotype is unknown.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN110770839B_ABST
    Figure CN110770839B_ABST
Patent Text Reader

Abstract

A computer method and system for quantifying a nucleic acid sample containing nucleic acids from one or more contributors is designed to: receive nucleic acid sequence reads obtained from the nucleic acid sample and mapped to alleles at a polymorphic locus; determine the allele counts for each allele at the polymorphic locus using the nucleic acid sequence reads; use a probabilistic mixture model that applies the probabilistic mixture model to the allele counts; and model the allele counts at the polymorphic locus using a probability distribution; use the probabilistic mixture model to quantify one or more fractions of nucleic acids from one or more contributors in the nucleic acid sample; determine the probability that a particular contributor among the one or more contributors has a particular genotype; and based on the posterior probability, call out that the nucleic acid sample includes nucleic acid from a particular contributor.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Patent Application No. 62 / 522,618, filed on June 20, 2017, the entire contents of which are incorporated herein by reference. Background Art

[0003] Sequencing data of a mixture of nucleic acids (e.g., DNA or RNA) from closely related genomes is often found in research as well as clinical settings, and quantification of the mixed contributors has been challenging when the original genomes are unknown. For example, in the context of microbiology and metagenomics, researchers and clinicians may need to quantify closely related bacterial strains of the same species in environmental samples. In a forensic setting, law enforcement officers may need to quantify and identify human individuals from a blood sample containing DNA from multiple individuals.

[0004] Another application is next-generation sequencing (NGS)-coupled liquid biopsy. NGS-coupled liquid biopsy is an emerging diagnostic strategy with potential applications in a variety of clinical settings. In the context of organ or tissue transplantation, NGS-coupled liquid biopsy provides a non-invasive method for monitoring the health of the allogeneic transplant by quantifying the amount of allogeneic DNA in the recipient's blood. In some applications, the donor and recipient genomes are unknown or partially unknown. SUMMARY OF THE INVENTION

[0006] Some embodiments proposed herein provide computer-implemented methods and systems for deconvolution of nucleic acid mixture samples, which nucleic acid mixture samples include nucleic acids of two or more contributors of unknown genotype. One aspect of the present disclosure relates to a method for quantitatively analyzing nucleic acid scores in nucleic acid samples, which nucleic acid samples include nucleic acids (e.g., DNA or RNA) of two or more contributors with different genomes. In some embodiments, as described below, nucleic acid mixture samples include biological tissues, cells, peripheral blood, saliva, urine and other biological fluids. In some applications, nucleic acid samples only include nucleic acids of a single contributor, and embodiments described herein can determine that nucleic acids of a single contributor account for 100% of nucleic acids in the sample. Therefore, although the description hereinafter refers to nucleic acid samples as nucleic acid mixture samples in some embodiments, it should be understood that samples can include nucleic acids of a single contributor, and the scores of contributors are 100% or 1. Of course, the method can also be used to quantify samples of nucleic acids comprising two or more contributors.

[0007] Because the various methods and systems provided herein implement the algorithms and processes using probability mixture models and Bayesian inference techniques, the embodiments provide technical improvements superior to conventional methods in deconvolution of nucleic acid (e.g., DNA or RNA) mixture samples. Some embodiments described herein refer to DNA samples, but it should be understood that the embodiments are also applicable to analyzing RNA samples. Some embodiments provide improved analytical sensitivity and specificity, provide more accurate deconvolution and quantification of nucleic acid mixture samples. Some embodiments allow accurate analysis of nucleic acid mixture samples with too low a nucleic acid amount that cannot accurately quantify contributor score or determine contributor genotype.

[0008] In some embodiments, the method is implemented on a computer system comprising one or more processors and a system memory, the computer system being configured to deconvolute a nucleic acid mixture sample comprising nucleic acids from two or more contributors.

[0009] Some embodiments provide a method for quantifying the nucleic acid fraction of a contributor in a nucleic acid mixture sample, the nucleic acid mixture sample comprising the nucleic acid of the contributor and at least one other contributor. The method involves: (a) receiving, by a computer system, a nucleic acid sequence read obtained from a nucleic acid sample and mapped to one or more alleles at one or more polymorphic loci; (b) using the nucleic acid sequence read and determining, by one or more processors, an allele count for each of the one or more alleles at the one or more polymorphic loci; (c) using a probabilistic mixture model, which applies the probabilistic mixture model to the allele counts and uses a probability distribution to model the allele counts at the one or more polymorphic loci, the probability distribution taking into account errors in the nucleic acid sequence read; (d) using the probabilistic mixture model and by one or more processors to quantify one or more fractions of nucleic acid of one or more contributors in the nucleic acid sample; (e) determining the probability that a particular contributor among the one or more contributors has a particular genotype; and (f) based on the posterior probability, calling out that the nucleic acid sample includes nucleic acid from a particular contributor.

[0010] In some embodiments, one or more contributors includes two or more contributors.

[0011] In some implementations, the method further includes determining a total number of contributors among the one or more contributors.

[0012] In some embodiments, one or more genotypes of one or more contributors are unknown. In some embodiments, the method further includes determining one or more allele configurations at each polymorphic locus of one or more polymorphic loci, each allele configuration comprising two or more allelic allele states of each contributor in one or more contributors. In some embodiments, the method further includes determining the estimated probability of one or more allele configurations.

[0013] In some embodiments, obtaining a posterior probability that a particular contributor among one or more contributors has a particular genotype comprises: (i) multiplying the prior probability of the genotype configuration by the likelihood of the genotype configuration; (ii) normalizing the product of (i) by the sum over the genotype space; and (iii) summing the genotype configurations that include the particular genotype to obtain the posterior probability.

[0014] In some embodiments, the specific genotype includes a multi-locus genotype, and the method further includes: summing the posterior probabilities that the contributor has the specific genotype at all loci among all contributors; and determining that the specified multiple loci genotypes occur in any contributor based on the summed probability. In some embodiments, the nucleic acid sample is a forensic sample, and the data of the multiple loci genotypes are obtained from an interested person, and the method further includes determining that the interested person is a contributor to the nucleic acid sample.

[0015] In some embodiments, the nucleic acid sample comprises DNA molecules and / or RNA molecules. In some embodiments, the nucleic acid sequence reads are obtained by sequencing the DNA molecules and / or RNA molecules using unique molecular indices.

[0016] In some embodiments, a probability distribution comprises a first binomial distribution. In some embodiments, a first binomial distribution is represented as follows:

[0017] n ij ~BN(n i , p ij )

[0018] n ij is the allele count of allele j at locus i; n i is the total allele count at locus i; and p ij is a probability parameter indicating the probability of allele j at locus i.

[0019] In some embodiments, the probability parameter p ijis a function of: (i) the fraction of nucleic acids in a nucleic acid sample that is one of the one or more contributors, or β; (ii) the genotype of the one or more contributors, or G; and / or (iii) errors in nucleic acid sequence reads, or θ.

[0020] In some embodiments, a probabilistic mixture model uses a Beta distribution to model errors in nucleic acid sequence reads. In some embodiments, a Beta distribution is defined by a mean parameter μ and a concentration parameter k. In some embodiments, the concentration parameter has a prior representing different noise conditions, and the concentration parameter varies across the locus.

[0021] In some embodiments, (c) comprises combining the first binomial distribution and the beta distribution to obtain n that follows a beta-binomial distribution. ii In some embodiments, the beta-binomial distribution has the following form:

[0022] BB(n ij |n i , μ, k).

[0023] In some embodiments, (c) comprises quantifying one or more fractions of nucleic acids of one or more contributors in a nucleic acid sample by maximizing a likelihood function of nucleic acid sequence reads. In some embodiments, (c) comprises: calculating a plurality of likelihood values ​​using a plurality of potential score values ​​and a likelihood function of allele counts determined in (b); identifying a potential score vector associated with a maximum likelihood value; and quantifying one or more fractions of nucleic acids of one or more contributors in a nucleic acid sample using the identified potential score vector.

[0024] In some embodiments, the likelihood function depends on P(G|π), which is the prior probability of the genotype of one or more contributors given the population allele frequency (π). In some embodiments, the prior probability P(G|π) is calculated using a marginal distribution that satisfies the Hardy-Weinberg equilibrium. In some embodiments, the prior probability is calculated by considering a dummy allele with a fixed prior probability of a representative mechanism dropout. In some embodiments, the probabilistic mixture model uses a second binomial distribution to model stutter errors in allele data. In some embodiments, the second binomial distribution is expressed as follows:

[0025] s ik ~BN(n i(k+1) , r i )

[0026] s ikis the Caton allele count at locus i of the Caton allele, which appears to be allele k but is actually caused by the Caton error of allele k+1; n i(k+1) is the raw allele count of allele k+1 at locus i; and r i is the stuttering rate of locus i.

[0027] In some embodiments, the jam rate r varies across loci and has a prior representing different noise conditions that is shared across loci. In some embodiments, (d) comprises quantifying the nucleic acid fraction of one or more contributors in a nucleic acid sample using a likelihood function that comprises the product of the likelihood of non-jamming allele counts and the likelihood of jamming allele counts. In some embodiments, (c) comprises adding a fixed number of molecules to the allele counts assigned to allele k+1 when determining the number of molecules that can potentially cause jams.

[0028] In some embodiments, the probability mixture model uses dummy out-of-sample alleles to model natural leakage. In some embodiments, the prior of the alleles outside the dummy sample is proportional to the number of unobserved alleles. In some embodiments, the number of unobserved alleles is estimated in the following manner: insert all integers between the shortest and longest integer value alleles observed, add any observed non-integer value alleles, and return the maximum result value and threshold value.

[0029] In some embodiments, (c) includes trimming genotype configuration from the data for quantitatively analyzing the nucleic acid fraction of one or more contributors in the nucleic acid sample. In some embodiments, trimming genotype configuration includes: by constructing a desired allele list and excluding loci that do not have enough contributors to explain all desired alleles, thereby limiting seemingly reasonable genotype configuration. In some embodiments, the list of desired alleles is essentially composed of the following alleles, which have allele counts that are higher than a threshold and too high to be non-seemingly reasonable due to the insertion of Caton. In some embodiments, the threshold is the sum of the following: (i) maximum non-Catton allele counts, and (ii) a value multiplied by the potential Caton donor allele counts. In some embodiments, trimming genotype configuration includes: removing genotype configurations that have poor matching between allele data and expected allele counts. In some embodiments, genotype configurations with poor matching have root mean square error (RMSE) values ​​greater than one or more thresholds.

[0030] In some embodiments, the alleles at one or more polymorphic loci comprise single nucleotide polymorphism (SNP) alleles and / or short tandem repeat (STR) alleles.

[0031] The disclosed embodiments also provide a computer program product including a non-transitory computer-readable medium on which program instructions for performing the recited operations and other computing operations described herein are provided.

[0032] Some embodiments provide a system for quantifying the nucleic acid fraction of a contributor in a nucleic acid mixture sample, the nucleic acid mixture sample comprising the nucleic acid of the contributor and at least one other contributor. The system includes: a sequencer for receiving nucleic acid from a test sample and providing nucleic acid sequence information from the sample; a processor; and one or more computer-readable storage media having stored thereon instructions for executing on the processor using the methods described herein.

[0033] One aspect of the present disclosure provides a computer system including a system memory and one or more processors. The processor is configured to: (a) receive a nucleic acid sequence read obtained from a nucleic acid sample and mapped to one or more alleles at one or more polymorphic loci; (b) using the nucleic acid sequence read, determine the allele count of each of the one or more alleles at the one or more polymorphic loci; and (c) using a probabilistic mixture model, which applies the probabilistic mixture model to the allele counts and uses a probability distribution to model the allele counts at the one or more polymorphic loci, the probability distribution taking into account errors in the nucleic acid sequence read; (d) using the probabilistic mixture model to quantify one or more fractions of nucleic acid of one or more contributors in the nucleic acid sample; (e) determine the posterior probability that a specific contributor in the one or more contributors has a specific genotype; and (f) based on the posterior probability, call out that the nucleic acid sample includes nucleic acid from the specific contributor.

[0034] In some embodiments, the system further comprises means for extracting nucleic acid from the nucleic acid sample.

[0035] In some implementations, the one or more processors are further configured to determine a total number of contributors among the one or more contributors.

[0036] In some embodiments, the one or more processors are further configured to determine an allelic configuration at each of the one or more polymorphic loci, the allelic configuration comprising allelic states of two or more alleles for each of the one or more contributors.

[0037] Another aspect of the present disclosure provides a non-transitory computer-readable medium storing program code that, when executed by one or more processors of a computer system, causes the computer system to implement a method for quantifying a nucleic acid sample comprising nucleic acid from one or more contributors, the program code comprising: (a) code for receiving nucleic acid sequence reads obtained from the nucleic acid sample and mapped to one or more alleles at one or more polymorphic loci; (b) code for determining, using the nucleic acid sequence reads, an allele count for each of the one or more alleles at the one or more polymorphic loci; (c) code for using a probabilistic mixture model that applies the probabilistic mixture model to the allele counts and uses a probability distribution to model the allele counts at the one or more polymorphic loci, the probability distribution taking into account errors in the nucleic acid sequence reads; (d) code for quantifying one or more fractions of nucleic acid from one or more contributors in the nucleic acid sample using the probabilistic mixture model; (e) code for determining a posterior probability that a particular contributor among the one or more contributors has a particular genotype; and (f) code for calling out that the nucleic acid sample includes nucleic acid from a particular contributor based on the posterior probability.

[0038] Although the examples herein relate to humans, and the language is primarily directed to humans, the concepts described herein are applicable to genomes from any plant or animal. These and other objects and features of the present disclosure will become more fully apparent from the following description and appended claims, or may be learned by the practice of the present disclosure as described below.

[0039] Incorporated herein by reference

[0040] All patents, patent applications, and other publications cited herein, including all sequences disclosed in such references, are expressly incorporated herein by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated for reference. All cited documents are, in relevant parts, fully incorporated herein by reference for the purposes indicated by the context of the references. However, the citation of any document should not be construed as an admission that it is prior art with respect to the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A-1C shows an overview of the bioinformatics algorithms and statistical models designed for contributor DNA quantification.

[0043] Figure 2A A block diagram is shown illustrating a process for quantifying one or more fractions of nucleic acid (eg, DNA or RNA) of one or more contributors in a nucleic acid sample.

[0044] Figure 2B A block diagram illustrating the components of a probabilistic mixture model is shown.

[0045] Figure 2C Schematic illustration of sequencing errors that convert one allele to another and a true allele to an unexpected allele.

[0046] Figure 3 A block diagram illustrating a process for evaluating a nucleic acid sample for nucleic acid containing one or more contributors is shown.

[0047] Figure 4 A block diagram of a typical computer system that may be used as a computing device according to some embodiments is shown.

[0048] Figure 5 One embodiment of a decentralized system for generating a call or diagnosis from a test sample is shown.

[0049] Figure 6 Options for performing various operations of some embodiments at different locations are shown.

[0050] Figures 7A-7F Results are shown using data obtained from actual DNA mixture samples to illustrate examples of some embodiments that can effectively quantify and deconvolute DNA mixture samples.

[0051] Figures 8A-8D Results are shown using simulated data to illustrate examples of some embodiments that can effectively quantify and deconvolute a DNA mixture sample. DETAILED DESCRIPTION

[0052] definition

[0053] Unless otherwise indicated, the practice of the methods and systems disclosed herein involves conventional techniques and apparatus commonly used in the fields of molecular biology, microbiology, protein purification, protein engineering, protein and DNA sequencing, and recombinant DNA within the skill of the art. Such techniques and apparatus are known to those skilled in the art and are described in many textbooks and reference works (see, e.g., Sambrook et al., "Molecular Cloning: A Laboratory Manual," 3rd ed. (Cold Spring Harbor,

[2001] ); and Ausubel et al., "Current Protocols in Molecular Biology"

[1987] ).

[0054] Numerical ranges include the numbers defining the ranges. Each maximum numerical limitation given throughout this specification is intended to include each lower numerical limitation, as if such lower numerical limitations were expressly written herein. Each minimum numerical limitation given throughout this specification will include each higher numerical limitation, as if such higher numerical limitations were expressly written herein. Each numerical range given throughout this specification will include each narrower numerical range that falls within such wider numerical range, as if such narrower numerical ranges were expressly written herein.

[0055] The headings provided herein are not intended to limit the disclosure.

[0056] Unless otherwise defined herein, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art. Various scientific dictionaries including the terms included herein are well known and available to those skilled in the art. Although any methods and materials similar or equivalent to those described herein can be used in the practice or testing of the embodiments disclosed herein, some methods and materials are described.

[0057] The terms defined immediately below are more fully described by reference to the entire specification. It should be understood that the disclosure is not limited to the specific methods, protocols and reagents described, as these may vary, depending on the context used by one skilled in the art. As used herein, the singular terms "a", "an" and "the" include plural references unless the context clearly indicates otherwise.

[0058] Unless otherwise indicated, nucleic acids are written left to right in 5' to 3' orientation and amino acid sequences are written left to right in amino to carboxyl orientation, respectively.

[0059] The term donor DNA (dDNA) refers to a DNA molecule derived from a transplant donor cell. In various embodiments, the dDNA is present in a sample obtained from a recipient who received a transplant tissue / organ from a donor. In some embodiments, the dDNA includes the following.

[0060] Circulating cell-free DNA or simply cell-free DNA (cfDNA) are DNA fragments that are not confined to cells and circulate freely in the bloodstream or other body fluids. cfDNA is known to have different sources, thus originating in some cases from tumor cells or cells affected by tumors, and in other cases from fetal cells of a fetus carried by a pregnant mother and circulating in the maternal blood. Typically, cfDNA is fragmented and includes only a small portion of the genome, which may be different from the genome of the organism from which the cfDNA was obtained.

[0061] The terms non-circulating genomic DNA (gDNA) or cellular DNA are used to refer to DNA molecules that are confined within a cell and typically include the entire genome.

[0062] The beta distribution is a family of continuous probability distributions defined on the interval [0, 1] that is parameterized by two positive shape parameters (e.g., denoted by α and β) that appear as exponentials of random variables and control the shape of the distribution. The beta distribution has been applied in various disciplines to model the behavior of random variables confined to intervals of finite length. In Bayesian inference, the beta distribution is the conjugate prior probability distribution of the Bernoulli, binomial, negative binomial, and geometric distributions. For example, the beta distribution can be used in Bayesian analysis to describe initial knowledge about the probability of success. If a random variable X follows a beta distribution, then the random variable X is written as X~beta(α, β).

[0063] The binomial distribution is a discrete probability distribution of the number of successes in a sequence of n independent experiments, each asking a yes / no question and each with its own Boolean outcome: a random variable containing a single bit of information: positive (with probability p) or negative (with probability q = 1-p). For a single trial, that is, n = 1, the binomial distribution is a Bernoulli distribution. The binomial distribution is often used to model the number of successes of a sample of size n drawn with replacement from a population of size N. If a random variable X follows the parameters and p∈[0,1], then the random variable X is written as X~B(n,p).

[0064] The Poisson distribution, denoted herein as Pois(), is a discrete probability distribution that expresses the probability of a given number of events occurring in a fixed time and / or space interval if these events occur at a known average rate and are independent of the time since the last event. The Poisson distribution can also be used for multiple events in other specified intervals, such as distance, area, or volume. According to the Poisson distribution, the probability of observing k events in an interval is given by the following equation:

[0065]

[0066] where λ is the average number of events in the interval or the event rate, also known as the rate parameter e is 2.71828, Euler's number, or the base of the natural logarithm, k is 0, 1, 2, … and k! is the factorial of k.

[0067] The gamma distribution is a two-parameter family of continuous probability distributions. There are three different parameter settings that are commonly used: shape parameter k and scale parameter θ; shape parameter α=k, and inverse scale parameter β=1 / θ, called the rate parameter; or shape parameter k and mean parameter μ=k / β. In each of these three forms, both parameters are positive real numbers. The gamma distribution is the maximum entropy probability distribution for a random variable X, for which E[X]=kθ=α / β is fixed and greater than zero, and E[ln(X)]=ψ(k)+ln(θ)=ψ(α)-ln(β) is fixed (ψ is the digamma function).

[0068] Polymorphism and genetic polymorphism are used interchangeably herein and refer to the occurrence of two or more alleles at a genomic locus in the same population, each allele having a distinct frequency.

[0069] Polymorphic site and polymorphic site are used interchangeably herein to refer to a locus on the genome where two or more alleles are located.

[0070] Allele frequency or gene frequency is the frequency of the allele of the gene (or variant of the gene) relative to other alleles of the gene, which can be expressed as a decimal or percentage.Allele frequency is usually associated with a specific genomic locus, because the gene is usually located at one or more loci.However, the allele frequency as used herein can also be associated with the fragment (size-based bin) based on size of DNA fragmentation.In this sense, the DNA fragmentation (such as cfDNA) comprising the allele is assigned to different fragments based on size.Relative to the frequency of other alleles, the frequency of the allele in the fragment based on size is the allele frequency.

[0071] The term "parameter" herein refers to a numerical value that characterizes an attribute of a system such as a physical feature, the value of which or other characteristics may affect relevant conditions such as a sample or DNA molecule. In some cases, the term "parameter" is used with reference to a variable that affects a mathematical relationship or model output, which may be an independent variable (i.e., an input to the model) or an intermediate variable based on one or more independent variables. Depending on the scope of the model, the output of one model may become the input to another model, thereby becoming a parameter of another model.

[0072] The term "plurality" refers to more than one element.

[0073] The term "paired-end reads" refers to reads from paired-end sequencing, which obtains one read from each end of a nucleic acid fragment. Paired-end sequencing can involve fragmenting a polynucleotide chain into short sequences called inserts. For shorter polynucleotides, such as cell-free DNA molecules, fragmentation is optional or unnecessary.

[0074] The terms "polynucleotide", "nucleic acid" and "nucleic acid molecule" are used interchangeably and refer to a covalently linked sequence of nucleotides (i.e., ribonucleotides for RNA and deoxyribonucleotides for DNA) in which the 3' position of the pentose of one nucleotide is linked to the 5' position of the pentose of another nucleotide via a phosphodiester group. Nucleotides include sequences of any form of nucleic acid, including but not limited to RNA and DNA molecules, such as cfDNA or cellular DNA molecules. The term "polynucleotide" includes but is not limited to single-stranded and double-stranded polynucleotides.

[0075] The term "test sample" herein refers to a sample generally derived from a biological fluid, cell, tissue, organ or organism, which contains nucleic acid or a mixture of nucleic acids. Such samples include, but are not limited to, sputum / oral fluid, amniotic fluid, blood, blood components or fine needle aspiration biopsy samples (e.g., surgical biopsy, fine needle aspiration biopsy, etc.), urine, peritoneal fluid, pleural fluid, etc. Although samples are generally taken from human subjects (e.g., patients), the assay can be used for samples of any mammal, including, but not limited to, dogs, cats, horses, goats, sheep, cattle, pigs, etc. The sample can be obtained directly from a biological source, or it can be pretreated to change the characteristics of the sample. For example, such pretreatment can include preparing plasma from blood, diluting viscous liquids, etc. Pretreatment methods can also include, but are not limited to, filtration, precipitation, dilution, distillation, mixing, centrifugation, freezing, freeze-drying, concentration, amplification, nucleic acid fragmentation, inactivation of interfering components, reagent addition, lysis, etc. If such pretreatment methods are applied to a sample, such pretreatment methods should generally result in the retention of the nucleic acid of interest in the test sample, sometimes at a concentration proportional to that in an untreated test sample (e.g., a sample that has not been subjected to any such pretreatment methods). With respect to the methods described herein, such "treated" or "processed" samples are still considered biological "test" samples.

[0076] The term "next generation sequencing (NGS)" herein refers to sequencing methods that allow for massively parallel sequencing of clonally amplified molecules and single nucleic acid molecules. Non-limiting examples of NGS include sequencing by synthesis using reversible dye terminators and sequencing by ligation.

[0077] The term "reading sequence" refers to a sequence obtained from a part of a nucleic acid sample. Generally, although not required, the reading sequence represents a short sequence of continuous base pairs in the sample. The reading sequence can be symbolically represented by the base pair sequence (represented by A, T, C or G) of the sample portion. It can be stored in a memory device and appropriately processed to determine whether it matches the reference sequence or meets other criteria. The reading sequence can be obtained directly from a sequencing device, or the reading sequence can be obtained indirectly from the stored sequence information related to the sample. In some cases, the reading sequence is a DNA sequence with sufficient length (e.g., at least about 25bp), which can be used to identify, for example, a larger sequence or region that can be compared and specifically assigned to a chromosome or genomic region or gene.

[0078] The term "genomic read" is used to refer to the read of any segment in the entire genome of an individual.

[0079] As used herein, the terms "aligned," "alignment," or "aligned" refer to the process of comparing a read or tag to a reference sequence and thereby determining whether the reference sequence contains the read sequence. If the reference sequence contains a read, the read can be mapped to the reference sequence, or in some embodiments, mapped to a specific position in the reference sequence. In some cases, the alignment simply indicates whether the read is a member of a specific reference sequence (i.e., whether the read exists or does not exist in the reference sequence). For example, an alignment of a read with a human chromosome 13 reference sequence will indicate whether the read exists in the reference sequence of chromosome 13. A tool that provides this information may be referred to as a set membership tester. In some cases, the alignment additionally indicates the position in the reference sequence to which the read or tag is mapped. For example, if the reference sequence is a full-length human genome sequence, the alignment may indicate that the read is present on chromosome 13, and may further indicate that the read is on a specific chain and / or site of chromosome 13.

[0080] The read sequence or label of the comparison is one or more sequences, which are identified as matching according to the order of their nucleic acid molecules and known sequences in the reference genome. Although usually realized by computer algorithms, the comparison can be performed manually because it is impossible to compare the read sequence within a reasonable period of time to realize the method disclosed herein. An example of an algorithm from comparing sequences is the efficient local alignment (ELAND) computer program for nucleotide data distributed as a part of the Illumina genomics analysis pipeline. Alternatively, a bloom filter or similar set membership tester can be used to make the read sequence and the reference genome comparison. Referring to U.S. Patent Application No. 61 / 552,374 filed on October 27, 2011, which is incorporated herein by reference as a whole. The matching of the sequence read sequence in the comparison can be 100% sequence matching or less than 100% (non-perfect matching).

[0081] The term "mapping" as used herein refers to the use of alignment or membership assignment to specifically assign sequence reads to a larger sequence, such as a reference genome, a subsequence of the larger sequence.

[0082] As used herein, the term "reference genome" or "reference sequence" refers to any specific known genome sequence, whether partial or complete, of any organism or virus that can be used to reference an identified sequence from a subject. For example, reference genomes for human subjects as well as many other organisms can be found at the National Center for Biotechnology Information (ncbi.nlm.nih.gov). "Genome" refers to the complete genetic information of an organism or virus expressed as a nucleic acid sequence.

[0083] In various embodiments, the reference sequence is significantly larger than the read sequence to which it is aligned. For example, it can be at least about 100 times larger, or at least about 1000 times larger, or at least about 10,000 times larger, or at least about 10,000 times larger. 5 times, or at least about 10 6 times, or at least about 10 7 times.

[0084] In one example, the reference sequence is a sequence of a full-length human genome. Such a sequence may be referred to as a genome reference sequence. In another example, the reference sequence is limited to a specific human chromosome, such as chromosome 13. In some embodiments, the reference Y chromosome is a Y chromosome sequence from human genome version hgl9. Such a sequence may be referred to as a chromosome reference sequence. Other examples of reference sequences include genomes of other species, and chromosomes, subchromosome regions (such as chains) of any species, etc.

[0085] In various embodiments, the reference sequence is a consensus sequence or other combination derived from multiple individuals. However, in certain applications, the reference sequence may be taken from a specific individual.

[0086] When used in the context of nucleic acids or nucleic acid mixtures, the term "derived" herein refers to the manner in which nucleic acids are obtained from the source from which they are derived. For example, in one embodiment, a mixture of nucleic acids derived from two different genomes refers to nucleic acids, such as cfDNA, which are naturally released by cells through naturally occurring processes (such as necrosis or apoptosis). In another embodiment, a mixture of nucleic acids derived from two different genomes refers to nucleic acids extracted from two different types of cells of a subject. For example, a nucleic acid mixture includes nucleic acids derived from donor cells and recipient cells obtained from an organ transplant subject. In some embodiments, a nucleic acid mixture comprises biological material from two or more contributor individuals. For example, a forensic sample comprising biological material from two or more individuals includes DNA from two or more individuals.

[0087] When used in the context of obtaining a specific quantitative value, the term "based on" herein means using another quantity as an input to calculate the specific quantitative value as an output.

[0088] The term "biological fluid" herein refers to a fluid obtained from a biological source, and includes, for example, blood, serum, plasma, sputum, lavage fluid, cerebrospinal fluid, urine, semen, sweat, tears, saliva, etc. As used herein, the terms "blood," "plasma," and "serum" expressly encompass fractions or processed portions thereof. Similarly, when a sample is obtained from a biopsy, swab, smear, etc., "sample" expressly encompasses processed fractions or portions obtained from the biopsy, swab, smear, etc.

[0089] As used herein, the term "corresponding to" sometimes refers to a nucleic acid sequence, such as a gene or chromosome, that is present in the genomes of different subjects and does not necessarily have the same sequence in all genomes, but is used to provide the identity of the sequence of interest (e.g., a gene or chromosome) rather than genetic information.

[0090] The term "contributor" herein refers to human contributors as well as non-human contributors, such as mammals, invertebrates, vertebrates, fungi, yeast, bacteria, and viruses. Although the examples herein involve humans, and the language is primarily directed to humans, the concepts disclosed herein are applicable to genomes from any plant or animal, and can be used in veterinary medicine, animal science, research laboratories, and the like.

[0091] As used herein, the term "sensitivity" refers to the probability that a test result is positive when the disease of interest is present. It can be calculated as the number of true positives divided by the sum of true positives and false negatives.

[0092] As used herein, the term "specificity" refers to the probability that a test result is negative when the disease of interest is not present. It can be calculated as the number of true negatives divided by the sum of true negatives and false positives.

[0093] As used herein, the term "primer" refers to an oligonucleotide that is separated as follows, and when placed under conditions (e.g., conditions including nucleotides, an inducing agent (such as a DNA polymerase) and a suitable temperature and pH) that induce the synthesis of an extension product, the oligonucleotide that is separated can serve as a synthesis starting point. For maximum amplification efficiency, the primer is preferably single-stranded, but can alternatively be double-stranded. If double-stranded, before being used to prepare an extension product, the primer is first treated to separate its chain. Preferably, the primer is an oligodeoxyribonucleotide. The primer must be long enough to initiate the synthesis of an extension product in the presence of an inducing agent. The exact length of the primer will depend on many factors, including temperature, primer source, use of the method, and parameters for primer design.

[0094] introduce

[0095] The present disclosure provides methods and systems for deconvoluting nucleic acid mixture samples including nucleic acids of two or more unknown genotype contributors, thereby providing various advantages and technical improvements. For example, some embodiments apply probabilistic mixture modeling, Bayesian inference techniques, and numerical optimization algorithms to quantify the contributor DNA in the mixture without knowing the genotype of the contributor.

[0096] In research and clinical settings, sequencing data of nucleic acid (e.g., DNA or RNA) mixtures from closely related genomes are often found, and when the original genome is unknown, it is always a challenge to quantify the mixed contributors. Attempts to deconvolve DNA mixtures have been made in the art, with limited success. Such attempts have been made using capillary electrophoresis (CE) allele data, which do not provide allele sequence information that may be useful in a clinical setting. In addition, analysis based on capillary electrophoresis is generally limited to relatively few allele numbers known in the database, and information outside those alleles cannot be captured. It is expected that the next generation sequencing technology is used to analyze DNA mixture samples. However, conventional methods for deconvoluting DNA samples have not yet been implemented to analyze NGS data. Even if people want to modify the conventional methods for NGS data analysis, the modification is not insignificant, and the success of such modifications is questionable. For example, the CE data of alleles are continuous, and the allele counts based on sequencing data are discrete. It will be appreciated by those skilled in the art that the model for continuous data does not work at all for discrete data, or will be performed suboptimally. Therefore, it is desirable to develop new methods for deconvolution of sequencing data (eg, NGS data) of DNA mixture samples.

[0097] Some embodiments provide methods and systems for quantifying contributor DNA from multi-marker targeted resequencing data of blood cfDNA or gDNA samples. Some embodiments provide methods and systems for quantifying contributor DNA from multi-marker targeted resequencing data of blood cfDNA or gDNA samples using novel probabilistic models and numerical optimization algorithms. Some embodiments provide methods and systems for quantifying contributor DNA of genetically related donors and recipients of unknown genotypes using Bayesian modeling with prior distributions encoding genetic relationships. By using genetic relationship information to provide prior information in a Bayesian framework, quantification of DNA mixtures can be improved compared to methods that do not use genetic relationship information.

[0098] Some embodiments provide methods and systems for estimating confidence intervals for DNA quantification by using the Cramer-Rao bound on the Hessian matrix of estimates of the log-likelihood function.

[0099] Allele bias in short sequencing read mapping confounds DNA quantification. In some embodiments, the confounding effect is reduced by using unbiased mapping of reads spanning variant sites.

[0100] Even if the genotypes of the contributing genome are completely unknown, the embodiments described herein can accurately estimate the contributor DNA fraction.The allele fractions of marker loci after PCR amplification can be reliably modeled with a Beta distribution.

[0101] Using an unbiased reference DNA sequence database, one can remove bias toward reference alleles and reliably estimate allele counts and sequencing errors at variant sites.

[0102] Embodiments described herein can employ a single sequencing run of a mixture DNA sample to estimate confidence intervals for predicted contributor DNA fractions.

[0103] Experimental pipeline

[0104] Figure 1 A-1C shows an overview of the bioinformatics algorithms and statistical models designed for contributor DNA quantification. Figure 1 A shows the experimental pipeline for sequencing based allogeneic DNA detection. Figure 1 B shows the unbiased read mapping workflow for allele counting. Figure 1 C shows the stratified probabilistic mixture model of allele counts per marker locus.

[0105] Some embodiments employ Figure 1 The experimental pipeline shown in A. The general experimental pipeline has the following steps.

[0106] 1) Obtain a blood sample containing DNA from two individuals.

[0107] 2) Depending on the application, extract the appropriate type of DNA, such as cellular DNA or cell-free DNA (cfDNA).

[0108] 3) Targeting and enriching specific variant sites or polymorphic sites of the genome by methods such as PCR amplification and hybridization. Variant sites are pre-selected to be variable in different populations of humans or bacteria. Alternatively, non-targeted whole genome sequencing can be performed and all variant sites will be covered.

[0109] 4) Sequencing the enriched DNA by NGS technology (such as some of the techniques described below) to obtain sequencing reads enriched for the target genomic region.

[0110] Formal Problem Statement

[0111] Formally, the Contributor DNA Quantification (CDQ) problem is as follows: Given sequencing data of a DNA sample containing two contributors, determine the score of each contributor in the sample. When the genotype of the contributor genome is unknown, the CDQ problem is called blind contributor DNA quantification (blind CDQ); the opposite is called non-blind CDQ. Some descriptions of some embodiments refer to the two contributors as donors and recipients, but they do not limit the application of the method to the organ donation environment. In some descriptions of some embodiments below, one or more contributors are equivalent to a donor, and another contributor is equivalent to a recipient.

[0112] Blind CDQ is a more difficult problem than unblinded CDQ, but it is broadly applicable to all scenarios where only a single sequencing experiment can be performed on a mixture sample, while unblinded CDQ requires a prior sequencing experiment to determine the genotypes of contributors (e.g., organ donors and recipients).

[0113] The computational approach described in this paper solves a blind CDQ problem, and components of the approach can be easily simplified or adapted to non-blind CDQ problems.

[0114] Calculation Method Overview

[0115] The blind CDQ calculation method has two main steps:

[0116] 1) Allele counting: A bioinformatics workflow for unbiased counting of sequencing reads for each allele at each target marker locus ( Figure 1 B), and

[0117] 2) Contributor DNA quantification: A hierarchical probabilistic model and an associated numerical optimization algorithm are used to quantify the contributor DNA fraction ( Figure 1 C).

[0118] Although some embodiments herein address only "relative quantification," meaning that the embodiment estimates the percentage or fraction of a DNA sample that originates from a contributor source, rather than the absolute amount (in terms of mass or copy number), if the total amount of input DNA is known, additional steps can be taken to convert relative abundance to absolute abundance.

[0119] Overview of the process used to quantify the fraction of contributors in a nucleic acid sample or to deconvolute a mixture of nucleic acid samples

[0120] Figure 2AA block diagram of a process 200 for quantifying one or more fractions of nucleic acid (e.g., DNA or RNA) of one or more contributors in a nucleic acid sample is shown. The nucleic acid sample includes a contributor and at least one other contributor's nucleic acid (e.g., DNA or RNA). The method is implemented on a computer system (such as the system described below) including one or more processors and a system memory. In some embodiments and applications, the description herein relates to DNA, but those skilled in the art will appreciate that RNA and other forms of nucleic acid may also be analyzed using the embodiments described herein. The various embodiments described herein may be used to analyze nucleic acid samples from nucleic acid of one or more contributors. In some embodiments, a method and system for quantifying one or more fractions of nucleic acid of one or more contributors is provided. In some descriptions herein, nucleic acid samples are referred to as mixture samples because samples may include nucleic acid from more than two contributors. However, it should be understood that the use of the term "mixture" indicates the possibility that the sample contains two or more contributor nucleic acids, and does not exclude the possibility that the sample only includes a single contributor nucleic acid. In the latter case, a number of 1 or a percentage of 100% (or a value within a margin of error) may be determined for a contributor.

[0121] In some embodiments, one or more contributors of the nucleic acid sample are included in the donor of the transplant in the recipient of the transplant. In some embodiments, the transplant comprises an allogeneic or xenogeneic transplant. In some embodiments, the nucleic acid sample is a biological sample obtained from the recipient. In some embodiments, the nucleic acid sample comprises cell-free nucleic acid. In some embodiments, the sample comprises cellular DNA.

[0122] Process 200 involves extracting nucleic acid molecules from a nucleic acid sample using techniques such as those described herein. See block 202.

[0123] Process 200 further includes amplification of the extracted nucleic acid molecules. Referring to frame 204. Various amplification techniques can be used, such as those described herein. In some embodiments, PCR is used to amplify the extracted nucleic acid molecules. In some embodiments, amplification is targeted at specific polymorphisms. In other embodiments, whole genome amplification can be performed, and allele data of specific polymorphic sites can be obtained by sequencing.

[0124] Process 200 also involves sequencing the amplified nucleic acid molecules using a nucleic acid sequencer to generate nucleic acid sequence reads. See block 206. Various sequencing techniques and equipment are further described below, which can be applied in operation 206.

[0125] Process 200 further relates to mapping the nucleic acid sequence reads to one or more polymorphic loci on the reference sequence. In some embodiments, alignment techniques can be used to map the nucleic acid sequence reads to one or more polymorphic loci. In other embodiments, unbiased mapping techniques can be used to match the nucleic acid sequence reads to the polymorphic loci. Referring to frame 208. In some embodiments, the nucleic acid sequence reads are mapped to specific alleles at the polymorphic loci. Meta-biased mapping techniques will be further described below. In some embodiments, one or more polymorphic loci (or polymorphic loci) include biallelic loci. In some embodiments, the alleles at one or more polymorphic loci include single nucleotide polymorphism (SNP) alleles.

[0126] In some embodiments, a unique molecular index (UMI) is attached to an extracted nucleic acid molecule, which is then amplified, sequenced, and mapped to a polymorphic locus or allele. The unique molecular index provides a mechanism to reduce errors that can occur during sample processing and analysis steps. For example, different reads that share the same unique molecular index (UMI) can be combined or overlapped to determine the sequence from which the read was derived, thereby effectively eliminating errors that occurred during sample processing and sequencing.

[0127] Process 200 further involves determining allele counts of nucleic acid sequence reads of alleles at one or more polymorphic loci using the method of nucleic acid sequence reads. See block 210.

[0128] Process 200 also involves applying the probability mixture model to the allele counts of the nucleic acid sequence reads. The probability mixture model uses probability distribution to model the allele counts of the nucleic acid sequence reads at one or more polymorphic loci. The probability distribution takes into account the errors in the nucleic acid sequence reads. The probability mixture model regards each allele count of the nucleic acid sequence reads as a random sample from the probability distribution.

[0129] In some embodiments, the probability distribution comprises a first binomial distribution. In some embodiments, the first binomial distribution comprises a number parameter indicating the total allele count at a locus and a possible parameter indicating the probability of a first allele at a locus. In some embodiments, the first binomial distribution is expressed as follows:

[0130] n ij ~BN(n i , p ij )

[0131] Where n ij is the allele count of nucleic acid sequence reads for allele j at locus i; n i is the total read count at locus i; and pij is a probability parameter indicating the probability of allele j at locus i. The allele probabilities of all possible alleles at this locus add up to 1.

[0132] In some embodiments, the probability parameter is a function of the nucleic acid fraction or β of the contributor. Possibly the parameter is also a function of the genotype of one or more contributors. The probability parameter is also a function of the error or λ caused by the sequencing operation of 206. In some embodiments, the probability parameter is obtained using the p1' values ​​in Table 3 described below. In some embodiments, the genotype of one or more contributors is unknown. In some embodiments, the probability mixture model includes Figure 2B The various probability distributions shown in .

[0133] return Figure 2A , process 200 summarizes quantifying one or more fractions of nucleic acid of one or more contributors in a nucleic acid sample using a probabilistic mixture model. See block 214. In some embodiments, quantification includes marginalizing over multiple possible combinations of genotypes to obtain a probability parameter. In some embodiments, quantification includes quantifying the nucleic acid fraction of a contributor using a likelihood function of allele counts of nucleic acid sequence reads determined in operation 210 conditioned on parameters of a probabilistic mixture model.

[0134] In some embodiments, quantitatively include using a plurality of potential score values ​​and the likelihood function of the allele count of the nucleic acid sequence reading to calculate a plurality of likelihood values.Quantitatively also relate to the potential score value associated with the likelihood value as the maximum in a plurality of likelihood values.In some embodiments, multiple likelihood values ​​are obtained for a plurality of parameters in the grid and their values.Quantitatively also relate to the nucleic acid score of the contributor in the nucleic acid sample with the potential score value with the maximum likelihood identified to be quantitative.In some embodiments, the likelihood function includes a plurality of marginal distributions for one or more polymorphic loci.

[0135] In some implementations, one or more contributors are modeled as two contributors, and the likelihood function is as follows:

[0136] L(n1,…n j |β,π)=П i ∑ g1jg2j BN(n ij , p ij (g 1j , g 2j ,λ,β))·P(g 1j , g 2j |π)

[0137] Among them, L(n1,…n j|β,π) is the observed allele counts n1 to n1 for alleles 1 to j given the parameters β (fraction of nucleic acid that is one of the contributors) and π (population allele frequency). j The likelihood of ij (g 1j , g 2j , λ, β) are probability parameters indicating the probability of the two contributor genotypes (g) based on allele j. 1j , g 2j ) is the probability of allele j at locus i; and P(g 1j , g 2j |π) is the a priori joint probability of observing the genotype of the two contributor alleles j given the population allele frequencies (π).

[0138] In some embodiments, the probabilistic mixture model takes into account errors caused by extracting nucleic acid molecules performed in 202 and errors caused by the sequencing operation in 206.

[0139] In some embodiments, the probabilistic mixture model uses a second binomial distribution to model the allele counts of the extracted nucleic acid molecules for alleles at one or more polymorphic loci. In some embodiments, the second binomial distribution is expressed as follows:

[0140] n ij ”~BN(n i ”,p ij )

[0141] Where n ij " is the allele count of the extracted nucleic acid molecule of allele j at locus i; n i ” is the total count of extracted nucleic acid molecules at locus i; and p ij is a probability parameter indicating the probability of allele j at locus i.

[0142] In some embodiments, the first binomial distribution is expressed as ij ” / n i " is the condition. In some embodiments, the first binomial distribution is reparameterized as follows:

[0143] n ij |n ij ”,n i ”~BN(n i , n ij ” / n i ”)

[0144] Where n ij is the allele count of nucleic acid sequence reads of allele j at locus i; n i" is the total number of nucleic acid molecules extracted at locus i, which is equal to the total genome copy number n"; n i is the total read count at locus i; and n ij ” is the number of extracted nucleic acid molecules of allele j at locus i.

[0145] In some embodiments, a probabilistic mixture model uses a first Beta distribution to approximate n ij In some embodiments, the first beta distribution has a mean and variance that match the mean and variance of the second binomial distribution.

[0146] In some embodiments, locus i is modeled as biallelic and a first Beta distribution is represented as follows:

[0147] n i1 ” / n”~Beta((n”-1)p i1 , (n”-1)p i2 )

[0148] where p i1 is a probability parameter indicating the probability of the first allele at locus i; and p i2 is a probability parameter indicating the probability of the second allele at locus i.

[0149] In some embodiments, the process includes combining a first binomial distribution and a first beta distribution to obtain n following the first beta binomial distribution. i1 The marginal distribution of .

[0150] In some embodiments, one or more contributors are modeled as two contributors and the first Beta-binomial distribution has the following form:

[0151] BB(n i1 , n i2 |n i , (n”-1)·p1(g 11 , g 21 ,λ,β),(n”-1)·p2(g 11 , g 21 , λ, β))

[0152] Where n i2 is the allele count of the nucleic acid sequence read of the second allele at locus i; p1(g 11 , g 21 , λ, β) are probability parameters indicating the first contributor genotype based on the first allele (g 11 ) and the second contributor genotype of the first allele (g 21) and the probability of the first allele due to sequencing error λ and contributor factor β; and p2(g 11 , g 21 , λ, β) are probability parameters indicating the first contributor genotype based on the first allele (g 11 ) and the second contributor genotype of the first allele (g 21 ) and the probability of the second allele due to sequencing error λ and contributor factor β.

[0153] In some embodiments, operation 214 includes quantifying one or more fractions of nucleic acid from one or more contributors in a nucleic acid sample using a likelihood function obtained using a first Beta-binomial distribution.

[0154] L(n1,n2|β,n”,λ,π)=Π i ∑ g11 g 21 BB(n i1 , n i2 |n i , (n”-1)·p1(g 11 , g 21 ,λ,β),(n”-1)·p2(g 11 , g 21 ,λ,β))·P(g 11 , g 21 |π)

[0155] where L(n1, n2|β, n", λ, π) is the likelihood of observing the allele count of the first allele (n1) and the allele count of the second allele (n2) given the parameters β, n", λ, and π; and P(g 11 , g 21 |π) is the first contributing genotype (g) of the first allele observed given the population allele frequency (π). 11 ) and the second contributor genotype of the first allele (g 21 ) is the prior joint probability of .

[0156] In some embodiments, operation 214 includes estimating the total number of extracted genome copies n" from the plurality of extracted nucleic acid molecules.

[0157] In some embodiments, the probabilistic mixture model takes into account errors caused by amplifying nucleic acid molecules performed in 204 and errors caused by sequencing operations in 206. In some embodiments, the amplification process is modeled as follows:

[0158] x t+1 =x t +y t+1

[0159] where x t+1 is the number of nucleic acid copies of a given allele after cycle t+1 of amplification; x t is the number of nucleic acid copies of a given allele after amplification cycle t; y t+1 is a new copy generated in cycle t+1, and it follows the binomial distribution y t+1 ~BN(x t , r t+1 ); and t+1 is the amplification rate at cycle t+1.

[0160] In some embodiments, the probabilistic mixture model uses a second Beta distribution to model the allele fractions of amplified nucleic acid molecules for alleles at one or more polymorphic loci. In some embodiments, locus i is modeled as a double allele, and the second Beta distribution is expressed as follows:

[0161] n i1 ' / (n i1 '+n i2 ')~Beta(n”·ρ i ·p i1 , n"·ρ i ·P i2 )

[0162] Where n i1 ' is the allele count of the amplified nucleic acid molecule of the first allele at locus i; n i2 ' is the allele count of the amplified nucleic acid molecule of the second allele at locus i; n" is the total extracted nucleic acid molecule count at any locus; p i is a constant related to the average amplification rate r; p i1 is the probability of the first allele at locus i; and p i2 is the probability of the second allele at locus i. In some embodiments, p i is (1+r) / (1-r) / [1-(1+r) -t In some embodiments, i Approximately (1+r) / (1-r).

[0163] In some embodiments, operation 214 includes combining the first binomial distribution and the second beta distribution to obtain n following a second beta-binomial distribution. i1 In some embodiments, the second beta-binomial distribution has the following form:

[0164] BB(n i1 , n i2 |n i , n"·ρi ·p i1 , n"·ρ i ·p i2 )

[0165] Where n i2 is the allele count of the nucleic acid sequence read of the second allele at locus i; p i1 is a probability parameter indicating the probability of the first allele at locus i; and p i2 is a probability parameter indicating the probability of the second allele at locus i.

[0166] In some embodiments, operation 214 includes reparameterizing the second Beta-binomial distribution by assuming that one or more polymorphic loci have the same amplification rate as:

[0167] BB(n i1 , n i2 |n i , (1+r) / (1-r)·p i1 (g 11 , g 21 ,λ,β),(1+r) / (1-r)·p i2 (g 11 , g 21 , λ, β))

[0168] where r is the amplification rate; and p i2 is a probability parameter indicating the probability of the second allele at locus i.

[0169] In some embodiments, operation 214 comprises quantifying one or more fractions of nucleic acid of one or more contributors in a nucleic acid sample using a likelihood function obtained using a second Beta-binomial distribution, the likelihood function being as follows:

[0170] L(n1,n2|β,r,λ,π)=

[0171] Π i ∑ g11 g 21 [BB(n i1 , n i2 |n i ,(1+r) / (1-r)·p1(g 11 , g 21 ,λ,β), (1+r) / (1-r)·p2(g 11 , g 21 ,λ,β))·P(g 11 , g 21 |π)]

[0172] where L(n1,n2|β,r,λ,π) is the likelihood of observing the allele count of the first allele (n1) and the allele count of the second allele (n2) given the parameters β, r, λ, and π.

[0173] In some embodiments, operation 214 comprises reparameterizing the second Beta-binomial distribution by defining the relative amplification rate of each polymorphic locus as proportional to the total reads per locus as:

[0174] BB(n i1 , n i2 |n i , c'·n i ·p i1 (g 11 , g 21 ,λ,β),c'·n i ·p i2 (g 11 , g 21 , λ, β))

[0175] where c' is the parameter to be optimized; and p i2 is a probability parameter indicating the probability of the second allele at locus i.

[0176] In some embodiments, operation 214 comprises quantifying one or more fractions of nucleic acid of one or more contributors in a nucleic acid sample using a likelihood function obtained using a second Beta-binomial distribution, the likelihood function being as follows:

[0177] L(n1,n2|β,c',λ,π)=

[0178] Π i ∑ g11g21 [BB(n i1 , n i2 |n i , c'·n i p1(g 11 , g 21 ,λ,β),c'·n i ·p2(g 11 , g 21 ,λ,β))·P(g 11 , g 21 |π)].

[0179] In some embodiments, the probabilistic mixture model takes into account errors caused by extracting nucleic acid molecules performed in 202 and amplifying nucleic acid molecules performed in 204, as well as errors caused by sequencing operations in 206. In some embodiments, the probabilistic mixture model uses a third beta distribution to model the allele portion of amplified nucleic acid molecules of alleles of one or more polymorphic loci, thereby accounting for errors caused by extracting nucleic acid molecules performed in 202 and amplifying nucleic acid molecules performed in 204, as well as sampling errors caused by sequencing operations in 206.

[0180] In some embodiments, locus i is modeled as biallelic and the third Beta distribution has the form:

[0181] n i1 ' / (n i1 '+n i2 ')~Beta([n”·(1+r i ) / 2-1]p i1 , [n”·(1+r i ) / 2-1]p i2 )

[0182] Where n i1 ' is the allele count of the amplified nucleic acid molecule of the first allele at locus i; n i2 ' is the allele count of the amplified nucleic acid molecule of the second allele at locus i; n" is the total nucleic acid molecule count extracted; r i is the amplification rate at locus i; p i1 is the probability of the first allele at locus i; and p i2 is the probability of the second allele at locus i.

[0183] In some embodiments, operation 214 includes combining the first binomial distribution and the third beta distribution to obtain n following the third beta-binomial distribution. i1 In some embodiments, the third beta-binomial distribution has the following form:

[0184] BB(n i1 , n i2 |n i , (n”·(1+r i ) / 2-1)·p1(g 11 , g 21 ,λ,β),(n”·(1+r i ) / 2-1)·p2(g 11 , g 21 , λ, β)).

[0185] In some embodiments, operation 214 comprises quantifying one or more fractions of nucleic acid of one or more contributors in a nucleic acid sample using a likelihood function obtained using a third Beta-binomial distribution, the likelihood function comprising:

[0186] L(n1,n2|β,n”,λ,π)=Π i ∑ g11g21 BB(n i1 , n i2 |n i , (n”·(1+r i ) / 2-1)·p1(g 11 , g 21 ,λ,β),(n”·(1+r i ) / 2-1)·p2(g 11 , g 21 ,λ,β))·P(g 11 , g 21 |π).

[0187] where L(n1, n2|β, n”, λ, π) is the likelihood of observing the allele count n1 of the first allele and the allele count n2 of the second allele given the parameters β, n”, λ, and π;

[0188] In some embodiments, process 200 further comprises estimating one or more confidence intervals for one or more fractions of nucleic acids of one or more contributors using the Cramer-Rao inequality.

[0189] In some embodiments, the mapping operation of 208 includes identifying reads in the nucleic acid sequence reads that match any sequence of a plurality of unbiased target sequences, wherein the plurality of unbiased target sequences include subsequences of a reference sequence and sequences that differ from the subsequences by a single nucleotide.

[0190] In some embodiments, the plurality of unbiased target sequences include five types of sequences: (i) reference target sequences that are subsequences of a reference sequence, each reference target sequence comprising a polymorphic locus and having a reference allele found on the reference sequence; (ii) alternative target sequences corresponding to the reference target sequences, each alternative target sequence having an alternative allele different from the reference allele found on the corresponding reference target sequence, wherein the corresponding sequences have the same length and the same position on the reference sequence; (iii) mutated reference target sequences, which comprise all possible sequences, each sequence differing from the reference target sequence by one nucleotide other than the nucleotide defining the difference between the reference allele and the alternative allele; and (iv) mutated alternative target sequences. targetsequences) comprising all possible sequences, each sequence differing from an alternative target sequence by only one nucleotide other than the nucleotide defining the difference between a reference allele and an alternative allele; and (v) unexpected allele target sequences corresponding to the reference target sequences, each unexpected allele target sequence having an unexpected allele different from a reference allele found on the corresponding reference target sequence and an alternative allele found on the corresponding alternative target sequence, wherein the corresponding sequences have the same length and the same position on the reference sequence.

[0191] In some embodiments, operation 208 comprises using the reads identified and the unbiased target sequences that they match to determine the allele counts of the nucleic acid sequence reads of the alleles at one or more polymorphic loci. In some embodiments, a plurality of unbiased target sequences comprise sequences that are truncated to have the same length as the nucleic acid sequence reads. In some embodiments, a plurality of unbiased target sequences comprise sequences stored in one or more hash tables, and the reads are identified using the hash tables.

[0192] In some embodiments, process 200 further includes determining an allelic configuration at each of the one or more polymorphic loci, each allelic configuration comprising allelic states of two or more alleles for each of the one or more contributors.

[0193] Figure 3 A block diagram illustrating a process 300 for evaluating a nucleic acid sample comprising nucleic acids of one or more contributors is shown. The process 300 begins by receiving nucleic acid sequence reads of one or more alleles at one or more polymorphic loci obtained from the nucleic acid sample. See block 302. In some embodiments, nucleic acid sequence reads are obtained by sequencing nucleic acids in the nucleic acid sample using various techniques described herein.

[0194] In some embodiments, a unique molecular index (UMI) is attached to an extracted nucleic acid molecule, which is then amplified, sequenced, and mapped to a polymorphic locus or allele. The unique molecular index provides a mechanism to reduce errors that can occur during sample processing and analysis steps. For example, different reads sharing the same unique molecular index (UMI) can be combined or overlapped to determine the sequence from which the read was derived, thereby effectively eliminating errors that occur during sample processing and sequencing. U.S. Patent Application No. 15 / 130,668, filed on April 16, 2016, and U.S. Provisional Patent Application No. 62 / 447,851, filed on January 18, 2017, describe various methods and systems for sequencing nucleic acids using unique molecular indexes, and the entire contents of each are incorporated herein by reference for all purposes.

[0195] Process 300 further involves determining allele counts for one or more alleles at one or more polymorphic loci using the nucleic acid sequence reads.

[0196] Process 300 also involves applying a probabilistic mixture model to allele counts. The probabilistic model uses a probability distribution to model the allele counts of the alleles at one or more polymorphic loci. The probability distribution illustrates errors in the allele data. The errors include errors derived from nucleic acid extraction, sample handling, and sequencing operations.

[0197] In some embodiments, probability distribution comprises the first binomial distribution.In some embodiments, the first binomial distribution comprises the parameter indicating the total allele count at the locus and the probability parameter indicating the probability of the first allele at the locus.In some embodiments, probability parameter is the function of the nucleic acid score of one or more contributors in the nucleic acid sample.Probability parameter is also the function of the genotype or G of one or more contributors, and the error or θ in the nucleic acid sequence reading data.In some embodiments, the error in the reading data comprises the error derived from nucleic acid extraction, sample handling and sequencing operation.

[0198] Process 300 also involves obtaining a likelihood value for the observed allele data given the model parameters and the potential nucleic acid score value. See block 308.

[0199] In some embodiments, process 300 involves using likelihood values ​​to quantify nucleic acid fractions of one or more contributors in a nucleic acid sample. See block 310 .

[0200] In some embodiments, process 300 further involves using the likelihood value to determine at least one genotype of at least one contributor. See block 312.

[0201] In some embodiments, the genotype of the contributor is unknown prior to process 300 .

[0202] In some embodiments, the probability mixture model uses a beta distribution to model the errors in the allele data. In some embodiments, the beta distribution is defined by a mean parameter and a concentration parameter. In some embodiments, the concentration parameter has a discrete prior representing different noise conditions. The concentration parameter changes on the locus.

[0203] In some embodiments, the quantification of operation 310 includes combining a first binomial distribution and a beta distribution to obtain a marginal distribution that follows a beta-binomial distribution.

[0204] In some embodiments, the quantification at 310 includes quantifying the nucleic acid fraction of one or more contributors in the nucleic acid sample using a likelihood function of the allele data. In some embodiments, quantification involves calculating multiple likelihood values ​​using a likelihood function of multiple potential score values ​​and allele counts. Quantification also involves identifying a potential score vector associated with the maximum likelihood value, and quantifying the nucleic acid fraction of one or more contributors in the nucleic acid sample using the identified potential score vector.

[0205] In some embodiments, the likelihood function depends on P(G|π), which is the prior probability of the genotype of one or more contributors given the population allele frequency (π). In some embodiments, a prior probability is calculated considering a dummy allele with a fixed prior probability representing a mechanism of missingness.

[0206] In some embodiments, one or more contributors comprise two or more contributors.In some embodiments, process 300 comprises the operation of determining the total number of contributors in one or more contributors.In some embodiments, one or more genotypes of one or more contributors are unknown, and process 300 comprises the operation of determining the allele configuration at each locus of one or more polymorphic loci, and this allele configuration comprises the allele state of two or more alleles of each contributor in the contributor.In some embodiments, process 300 comprises the operation of determining the estimated probability of allele configuration.

[0207] In some embodiments, process 300 further includes obtaining a posterior probability that a particular contributor among one or more contributors has a particular genotype. In some embodiments, process 300 further includes calling out that the nucleic acid sample includes nucleic acid from a particular contributor based on the posterior probability. In some embodiments, obtaining a posterior probability that a particular contributor among one or more contributors has a particular genotype includes: (i) multiplying the prior probability of the genotype configuration by the likelihood of the genotype configuration; (ii) normalizing the product of (i) by the sum over the genotype space; and (iii) summing the genotype configurations that include the particular genotype to obtain the posterior probability.

[0208] In some embodiments, the specific genotype includes a multi-locus genotype, and the method further includes: summing the posterior probabilities that the contributor has the specific genotype at all loci among all contributors; and determining that the specific multi-locus genotype occurs in any contributor based on the summed probability.

[0209] In some embodiments, the nucleic acid sample is a forensic sample and the data of the genotypes of the plurality of loci is obtained from a person of interest. The process further includes determining that the person of interest is a contributor to the nucleic acid sample.

[0210] In some embodiments, a probabilistic mixture model uses a second binomial distribution to model stuttering errors in allele data. In some embodiments, the second binomial distribution is expressed as follows:

[0211] s ik ~BN(n i(k+1) , r i )

[0212] where s ik is the Caton allele count at locus i of the Caton allele, which appears to be allele k but is actually caused by the Caton error of allele k+1; n i(k+1) is the raw allele count of allele k+1 at locus i; and r i is the stutter rate at locus i.

[0213] In some embodiments, the jam rate r varies across loci and has a prior that represents different noise conditions that is shared across the loci.

[0214] In some embodiments, operation 310 includes quantifying the nucleic acid fraction of one or more contributors in a nucleic acid sample using a likelihood function that includes the product of a likelihood of a non-catton allele count and a likelihood of a catton allele count.

[0215] In some embodiments applying a probabilistic mixture model comprises adding a fixed number of molecules to the allele count assigned to allele k+1 when determining the number of molecules that could potentially be derived therefrom.

[0216] In some embodiments, the probabilistic mixture model uses virtual out-of-sample alleles to model natural dropouts. In some embodiments, the prior of the virtual out-of-sample alleles is proportional to the number of unobserved alleles. In some embodiments, the number of unobserved alleles is estimated in the following manner: interpolate all integers between the shortest and longest observed integer-valued alleles, add any observed non-integer-valued alleles, and return the maximum and standard values ​​of the result values.

[0217] In some embodiments, application probability mixture model relates to pruning genotype configuration from the data for quantitative nucleic acid fraction of one or more contributors in nucleic acid sample.In some embodiments, pruning genotype configuration relates to: by constructing the list of required alleles and excluding the loci that do not have enough contributors to explain all required alleles, thereby limiting the seemingly reasonable genotype configuration.In some embodiments, the list of required alleles is basically composed of the following alleles, which have higher than threshold value and too high to non-seemingly reasonable allele counts due to Stutter insertion.In some embodiments, the threshold value is the sum of the following: (i) maximum non-Stutter allele count, and (ii) the value multiplied by the potential Stutter donor allele count.In some embodiments, pruning genotype configuration relates to removing the genotype configuration with poor matching between allele data and expected allele count.In some embodiments, the genotype configuration with poor matching has a root mean square error (RMSE) value greater than one or more threshold values.

[0218] In some embodiments, the alleles at one or more polymorphic loci comprise single nucleotide polymorphism (SNP) alleles and / or short tandem repeat (STR) alleles.

[0219] Method for unbiased mapping of reads to landmark sites

[0220] Conventional computational methods for mapping nucleic acid (e.g., DNA or RNA) sequencing reads to genomes can be biased by the reference genome used. Since there is only one allele (the reference allele) per variant site in the reference genome, mismatches between reads and the reference are treated as sequencing errors in existing read mapping algorithms. The problem is that when reads containing non-reference alleles are treated as containing sequencing errors, the alignment confidence (score) is reduced, and therefore, they are less likely to be retained as credibly mapped reads in subsequent filtering steps. This mapping bias can skew allele counts ( Figure 1 B), and subsequently impairs the estimate of the contributor DNA fraction.

[0221] To address the mapping bias issue and achieve optimal CDQ, some embodiments provide a novel workflow for mapping reads to variant sites. The new read mapping method can perform unbiased allele counting and estimate sequencing errors for variant and non-variant sites.

[0222] The read mapping workflow is as follows. The workflow first generates five types of sequences based on 1) the reference sequence and 2) the known alleles of the variant site (see Table 1). If more than one single mutation is allowed per sequence, more types of sequences will be generated. The five types of sequences are called ref, alt, ref.mut, alt.mut, and snp.mut. For example, for each biallelic SNP marker site covered by a target sequence of length L, there is one ref, one alt, [L-1]x3ref.mut, [L-1]x3 alt.mut, and 2 snp.mut sequences. All five types of sequences are then included in the "unbiased target sequence" database ( Figure 1 B). Then, depending on the length of the read from the sequencer, the unbiased target sequence is truncated into two versions. Let r be the read length. Version 1 of the truncated target sequence includes the r 5' bases of all unbiased target sequences, and version 2 of the truncated target sequence includes the reverse complement of the r 3' bases of all unbiased target sequences. The redundant sequences in the truncated target sequence are then removed. The unique sequences in the two truncated sequence databases are then recorded in two hash tables. Next, the sequence reads are counted using the hash tables. For the paired end sequencing strategy, the R1 reads and the R2 reads are counted using the first hash table and the second hash table, respectively. For non-paired end sequencing, all reads are counted using the first hash table. Finally, for each marker site, the counts are summarized into the five types defined above, depending on whether the truncated unbiased target sequence corresponds to the type in Table 1.

[0223] A similar strategy can be implemented when a sequence alignment tool is used instead of using a hash table for mapping. For each marker site, ref and alt types of sequences are generated to form an unbiased sequence database. Each sequencing read is then aligned to this database with up to a predefined number of sequencing errors. The mapped reads are then classified based on Table 1. For SNP markers, only biallelic SNPs are considered here.

[0224] Table 1. Definitions of the five types of sequences generated from the reference sequences around the variant sites.

[0225] type definition ref SNP loci using reference alleles alt SNP sites with alternative alleles ref.mut Single mutation at non-SNP site when SNP site is ref alt.mut Single mutation at non-SNP site when SNP site is alt snp.mut SNPs that use neither reference alleles nor alternative alleles

[0226] The proposed read mapping workflow addresses the read mapping bias issue when tested with real data. In the workflow, the observed reference-to-substitution error and substitution-to-reference error ratios are identical. The sequencing error rates at non-variant sites on the reference DNA copy and at non-variant sites on the alternative DNA copy are also identical.

[0227] Relating contributor DNA fractions to allele fractions

[0228] Assuming there are no sequencing errors

[0229] We assume that there are n d Donor cells and n r Based on these cells, embodiments define the fraction of minor contributors as β = n d / (n d +n r Depending on the donor and recipient genotypes at each specific locus, the two alleles have different scores (see Table 2 for details), and the general formula for calculating them is p1 = [g 11 (1-β)+c·g 21 ·β] / 2, and p2=[g 12 (1-β)+g 22 ·β] / 2. Note that g 11 and g 12 is the recipient genotype, i.e., the copies of alleles 1 and 2 in the recipient genome; g 21 and g 22 is the donor genotype, i.e. the copies of alleles 1 and 2 in the donor genome.

[0230] Table 2: Binomial model parameters p1 and p2 for the nine possible genotype combinations between donor and recipient pairs at a given variant site

[0231] <![CDATA[g 11 ]]> <![CDATA[g 21 ]]> <![CDATA[p1]]> <![CDATA[p2]]> 0 0 0 1 0 1 β / 2 1-β / 2 0 2 β 1-β 1 0 (1-β) / 2 (1+β) / 2 1 1 1 / 2 1 / 2 1 2 (1+β) / 2 (1-β) / 2 2 0 1-β β 2 1 1-β / 2 β / 2 2 2 1 0

[0232] Modeling sequencing errors

[0233] When there are two known alleles at a variant site, sequencing errors convert one allele to the other in addition to converting the two known alleles to the two remaining nucleotides at that locus. The result is that the allele fraction in the sequencing read will be different from the allele fraction in the NGS input DNA sample.

[0234] Figure 2C Schematically illustrates sequencing errors that convert one allele to another and true alleles to unexpected alleles. Panel (A) shows nucleotide-dependent sequencing errors, and panel (B) shows uniform sequencing errors.

[0235] Let N1, N2 be the nucleotides of allele 1 and allele 2. Let p1′, p2′ be the probability of observing a read of allele 1 and allele 2, respectively, whether it is real or due to sequencing error; and p0′=1-p1′-p2′ is the probability of observing two unexpected alleles due to sequencing error. Let λ N1N2 is the mutation rate (probability) from N1 to N2, where N1 and N2 are unique for each SNP site, and

[0236] λ N1# : The probability of mutation from N1 to any of the non-N1 nucleotides within 3 nucleotides.

[0237] Figure 2C The transition diagram between the 4 nucleotides of the SNP site is shown. Based on this, the embodiment obtains the following equation for converting from the true allele fractions p1, p2 to the observed allele fractions p1', p2' and p0':

[0238] p1′=p1-p1·λ N1# +p2·λ N2N1

[0239] p2′=p2-p2λ N2#+ p1·λ N1N2

[0240] p0′=p1·(λ N1#- λ N1N2 )+p2·(λ N2#- λ N2N1 ).

[0241] When embodiments assume a uniform sequencing error rate that is independent of nucleotide identity, embodiments have

[0242] p1′=p1·(1-3·λ)+p2·λ

[0243] p2′=p2·(1-3·λ)+p1·λ

[0244] p0′=2λ.

[0245] When implementation ignores unexpected alleles

[0246] p1′=(p1·(1-3·λ)+p2·λ) / (1-2λ)

[0247] p2′=(p2·(1-3·λ)+p1·λ) / (1-2λ),

[0248] Using o(λ 2 ) approximation error, which can be rewritten as

[0249] p1′=p1·(1-λ)+p2·λ

[0250] p2′=p2·(1-λ)+p1·λ

[0251] Table 3 lists the formulas that relate the contributor fraction β to the observed allele fraction p1′, depending on the contributor's genotype.

[0252] Table 3: Expected probability p of observing alleles 1 and 2 after accounting for sequencing errors i1 (g 11 , g 21 ,λ,β) and p i2 (g 11 , g 21 ,λ,β), conditional on each donor / recipient genotype combination. Here, it is assumed that for all N1 and N2,λ N1N2 =λ. Since the mutation rate λ is small, a first-order approximation is used.

[0253] <![CDATA[g 11 ]]> <![CDATA[g 21 ]]> <![CDATA[p1′]]> <![CDATA[p2′]]> 0 0 λ 1-λ 0 1 β / 2+λ-βλ 1-β / 2-λ+βλ 0 2 β+λ-2βλ 1-β-λ+2βλ 1 0 (1-β) / 2+βλ (1+β) / 2-βλ 1 1 1 / 2 1 / 2 1 2 (1+β) / 2-βλ (1-β) / 2+βλ 2 0 1-β-λ+2βλ β+λ-2βλ 2 1 1-β / 2-λ+βλ β / 2+λ-βλ 2 2 1-λ λ

[0254] Overview of DNA extraction, PCR (amplification), and sequencing models

[0255] Three probability models are provided ( Figure 1 C) to simulate a general experimental pipeline ( Figure 1 A) has three main components: 1) DNA extraction; 2) DNA amplification (e.g., PCR) as a method to enrich the target DNA; 3) sequencing (e.g., NGS sequencing).

[0256] The following symbols are used in the mathematical model detailed in Table 4.

[0257] p1, p 1i : The probability of allele 1 at locus i. Note that when the implementation focuses on a single locus, the subscript i is omitted.

[0258] p1, p 2i : The probability of allele 2 at locus i

[0259] n1,n 1i , n2, n 2i : Read counts of allele 1 and allele 2 at locus i

[0260] n,n i =n 1i +n 2i : Total read counts for the two known alleles of locus i

[0261] g 11 , g 12 , g 11i , g 12i : The recipient genotype, i.e. the copies of alleles 1 and 2 in the recipient genome

[0262] g 21 , g 22 , g 21i , g 22i : Donor genotype, i.e. the number of copies of alleles 1 and 2 in the donor genome

[0263] B(): Beta function

[0264] Beta(), BN(), Pois(), Gamma(): Beta, binomial, Poisson, and gamma distributions

[0265] N: the number of cells providing the DNA in the sample

[0266] β: donor DNA fraction, which is defined as the percentage of donor-derived DNA in the sample.

[0267] n r =N·(1-β), n d = N·β: the number of recipient cells and donor cells that provide the DNA in the sample

[0268] Table 4: Statistical models for the three main components of the general experimental pipeline. The model for each component is conditioned on the previous component. Models are conditioned on each locus.

[0269]

[0270] DNA Extraction Model

[0271] When cfDNA or cellular DNA is extracted from a blood sample, the DNA obtained is a small sample from a large DNA pool, and therefore, the implementation modeled the counts of the two alleles at each locus as two Poisson distributions. Therefore, the DNA copies of allele 1 (n1") at the locus conditioned on the total count n" follow a binomial distribution: n1"~BN(n",p1), with mean μ0=n"·p1, and variance δ0 2 =n”·p1·p2. When the donor fraction β<0.2, δ0 2 ≈μ0.

[0272] When gDNA is extracted from the sample, the resulting gDNA amount for each locus can again be variable due to extraction losses. Considering p1 as the number of allele 1 in the input sample, the amount of allele 1 in the extracted DNA can again be modeled by a binomial distribution: n1"~BN(n",p1).

[0273] PCR amplification model

[0274] We model the PCR amplification process as a random process to obtain the probability distribution of the allele 1 count in the PCR product. Let x t is the number of DNA copies of a given allele after cycle t of PCR amplification, let r t is the amplification rate of cycle t, and let y t is the new copy generated at cycle t. By assuming that each DNA strand has a probability r of being amplified and added to the DNA pool t , this embodiment has the following expansion model:

[0275] x t+1 =x t +y t+1 , where y t+1 ~BN(x t , r t+1 ) follows a binomial distribution, where x t and r t+1 as a parameter.

[0276] Based on this model, embodiments require that the number of DNA copies of a locus in a PCR product approximately follows a gamma distribution. The reasons are provided below.

[0277] Step 1: Use the Yule process (continuous-time random process) to approximate the PCR (discrete-time random process).

[0278] PCR processx t+1 =x t +y t+1 , where y t+1 ~BN(x t , r t+1) is a discrete-time pure birth process: in a given time t cycle, each copy of the DNA is generated at a certain rate r t "born" independently. The continuous-time version of the pure birth process is known as the Yule-Furry process. For the continuous-time birth process, it is known that the final number of copies of a locus at a given time t follows a negative binomial distribution. When the number of PCR cycles is not close to 1, embodiments can use the same distribution to approximate the discrete-time birth process.

[0279] Step 2: Use the gamma distribution (continuous distribution) to approximate the negative binomial distribution (discrete distribution).

[0280] The negative binomial random variable (rv) can be written as the sum of iid geometric rvs. It is known that the exponential distribution is a continuous form of the geometric distribution. Therefore, the sum of iid exponential rvs following the gamma distribution is a continuous form of the sum of binomial rvs as the negative binomial.

[0281] The following embodiment estimates the parameters of the gamma distribution of allele counts in PCR products.

[0282] Based on the total variance law var(x t+1 )=var(E(x t+1 |x t )+E(var(x t+1 |x t ), the implementation method can derive x t The mean and variance of are as follows:

[0283] μ t+1 =μ t ·(1+r t+1 )

[0284] δ t+1 2 =μ t ·r t+1 ·(1-r t+1 )+δ t 2 ·(1+r t+1 ) 2

[0285] Among them, μ t =E(x t ), δ t 2 =var(x t ).

[0286] Assuming the average amplification rate per PCR cycle is r t+1 = r, then the implementation method has

[0287] μ t =μ0·(1+r) t

[0288] δ t 2 =μ0·(1+r) t ·[(1+r) t -1]·(1-r) / (1+r)+δ0 2 (1+r) 2t

[0289] Note that μ0 and δ0 2 are the mean and variance of the DNA allele counts in the PCR amplification input, and they can be calculated based on the DNA extraction model described above. Alternatively, if the embodiment does not treat the cfDNA / cellular DNA allele counts as random variables, the embodiment has μ0=n1″ or n2″, and δ0 2 =0.

[0290] The corresponding gamma distribution G(x) matches this mean and variance t |k,θ)=x k-1 e -x / θ / [θ k ·Γ(k)] has parameters:

[0291] θ=[(1+r) t -1]·(1-r) / (1+r)+δ0 2 / μ0·(1+r) t

[0292] k=μ0·(1+r) t / [[(1+r) t -1]·(1-r) / (1+r)+δ0 2 / μ0·(1+r) t ].

[0293] For a given locus with two alleles and two initial copies (n1', n2'), assuming the same amplification rate r1 = r2 = r for the two alleles at each locus, the two corresponding gamma distributions G(n1'|k1, θ1) and G(n2'|k2, θ2) have the following parameters:

[0294] θ1=[(1+r) t -1]·(1-r) / (1+r)+P2·(1+r) t

[0295] θ2=[(1+r) t -1]·(1-r) / (1+r)+p1·(1+r) t

[0296] k1=n″p1 / [[1-(1+r) -t ]·(1-r) / (1+r)+p2]

[0297] k2=n″p2 / [[1-(1+r) -t ]·(1-r) / (1+r)+p1].

[0298] When the embodiment adjusts the PCR model based on the DNA extraction model, stμ0=n1″ or n2″ and δ0 2 = 0, the implementation method has

[0299] θ1=[(1+r) t -1]·(1-r) / (1+r)

[0300] θ2=[(1+r) t -1]·(1-r) / (1+r)

[0301] k1=n l ″·(1+r) / (1-r) / [1-(1+r) -t ]

[0302] k2=n2″·(1+r) / (1-r) / [1-(1+r) -t ].

[0303] Therefore, the allele copies n1′ and n2′ in the PCR products follow two gamma distributions with the same scale parameters θ1 and θ2, which depend only on the PCR process (number of cycles and amplification rate).

[0304] n1′ / (n1′+n2′)~Beta(n1″·ρ, n2″·ρ),

[0305] Where, ρ = (1 + r) / (1 - r) / [1 - (1 + r) -t ], or about ρ = (1 + r) / (1 - r) when the number of cycles t is large, is a constant related to the amplification rate r, which depends only on the PCR process. For a specific locus, it is written as n i1 ′ / (n i1 ′+n i2 ′)~Beta(n i1 ″·ρ i , n i2 ″·ρ i ) to capture locus-specific PCR amplification rates.

[0306] If the implementation ignores DNA sampling and assumes that all loci have the same total DNA copy number n i″=n″, then n i1 ″=n″·ρ i ·p i1 And n i2 ″=n″·ρ i ·p i2 The allele fractions of the loci in the PCR products are as follows:

[0307] n i1 ′ / (n i1 ′+n i2 ′)~Beta(n″·ρ i ·p i1 , n″·ρ i ·p i2 ).

[0308] Note that without the gamma distribution approximation, the implementation has n1′~NB(r1,p) and n2′~NB(r2,p), and the ratio n1′ / (n1′+n2′) does not have a closed form distribution. For the gamma distribution approximation, n1′~Gamma(n1″·ρ,θ) and n2′~Gamma(n2″·ρ,θ), and n1′ / (n1′+n2′) follows a Beta distribution.

[0309] Sequencing Model for Read Counts

[0310] NGS sequencing is the process of sampling from a pool of DNA molecules provided to a sequencer and reading out the sequences of these molecules. The fraction of allele 1 at locus i in the PCR product is n i1 ′ / (n i1 ′+n i2 ′). This fraction determines the probability of allele 1 read appearing in the sequencing result. The total number of reads per locus n i As the condition, n i1 The distribution of the read count of allele 1 of the locus is then modeled as a binomial distribution n i1 ~BN(n i , n1′ / (n1′+n2′)).

[0311] Modeling genetic correlations between contributors as a prior distribution

[0312] If the contributor (donor / recipient) genotypes are fully known, they can be directly incorporated (using Table 2 or Table 3) as parameters of the above component models. However, when the genotypes are unknown, embodiments can utilize information on the genetic relationship between the donor and recipient, which is typically available in clinical applications.

[0313] We represent different types of donor-recipient relationships as different prior distributions over the space of possible genotype combinations of the donor and recipient. Assuming Hardy-Weinberg equilibrium, the genotype distribution of a single individual is P(g 母 =[0, 1, 2]) = [(1-π) 2 , 2π(1-π), π 2 ], assuming that the population frequency of allele 2 is π. Note that all genetic correlations are the result of parent-offspring relationships. Based on the genetic relationships between parents and children for a given biallelic marker site (Table 5), the implementation method can calculate the joint distribution of any genetic relationship.

[0314] Table 5: Probability distribution of offspring genotypes given the genotypes of the parents at a given locus, and the joint distribution between the parents assuming that the parents are not related.

[0315] <![CDATA[g 父 ]]> <![CDATA[G 母 ]]> The probability of having children with genotype [0, 1, 2] <![CDATA[P(g 父 ,g 母 )]]> 0 0 [1,0,0] <![CDATA[(1-π) 4 ]]> 0 1 [1 / 2,1 / 2,0] <![CDATA[2π(1-π) 3 ]]> 0 2 [0,1,0] <![CDATA[π 2 (1-π) 2 ]]> 1 0 [1 / 2,1 / 2,0] <![CDATA[2π(1-π) 3 ]]> 1 1 [1 / 4,1 / 2,1 / 4] <![CDATA[4π 2 (1-π) 2 ]]> 1 2 [0,1 / 2,1 / 2] <![CDATA[2π 3 (1-π)]]> 2 0 [0,1,0] <![CDATA[π 2 (1-π) 2 ]]> 2 1 [0,1 / 2,1 / 2] <![CDATA[2π 3 (1-π)]]> 2 2 [0,0,1] <![CDATA[π 4 ]]>

[0316] The following are the prior distributions for the following types of genetic relationships: parent-child, child-parent, sibling, uncle / aunt-nephew, nephew-uncle / aunt, and unrelated.

[0317] Joint distribution between father and offspring genotypes

[0318] As an example, the joint distribution of father-child donor-recipient genotypes (GT) is calculated using the following formula:

[0319] P(recipient = my GT, donor = father GT) = ∑ 母亲 GT[P(my GT|father GT, mother GT)·P(father GT, mother GT)],

[0320] Among them, the values ​​of P(my GT|father GT, mother GT) and P(father GT, mother GT) are taken from the 3rd and 4th columns of Table 5 respectively.

[0321] Joint distribution between sibling genotypes

[0322] As an example, based on the conditional independence of the two sibling genotypes given the parental genomes, the following formula is used to calculate the joint distribution of my-sibling donor-recipient genotypes:

[0323] P(recipient = me GT, donor = sibling GT) = ∑ 母亲GT ∑ 父亲GT [P(I GT|Father GT, Mother GT)·P(Siblings GT|Father GT, Mother GT)·P(Father GT, Mother GT)],

[0324] Among them, the values ​​of P(my GT|father GT, mother GT), P(siblings GT|father GT, mother GT) and P(father GT, mother GT) are taken from the third, third and fourth columns of Table 5, respectively.

[0325] Joint distribution between uncle-nephew genotypes

[0326] As an example, the uncle / aunt-nephew / niece donor-recipient genotype joint distribution is calculated using the following formula:

[0327] P(recipient = me GT, donor = uncle GT)

[0328] =∑ 祖母GT ∑ 祖父GT ∑ 母亲GT ∑ 父亲GT [P(I GT|father GT, mother GT)·P(mother GT)·P(father GT|grandfather GT, grandmother GT)·P(uncle GT|grandfather, grandmother GT)·P(grandfather GT, and mother GT)]

[0329] =∑ 母亲GT ∑ 父亲GT P(my GT|father GT, mother GT)·P(mother GT)·P(father GT, uncle GT),

[0330] Among them, the value of P(my GT|father GT, mother GT) is taken from the third column of Table 5, and the value of P(father GT, uncle GT) is the same as P(recipient = my GT, donor = sibling GT).

[0331] Table 6 summarizes the results of the above derivation, and a specific example for a given population SNP allele frequency value of π = 0.5 is provided in Table 7. Additional relationships, such as grandparent-grandson, grandson-granddaughter, half-siblings, and cousins, can be derived based on the same basic principles.

[0332] Table 6: Prior distribution of related or unrelated genomes P(g 11 , g 21 ). Assumptions 1) All SNPs are from autosomes and 2) All married couples are genetically unrelated. 11 is the recipient genome, g 21 It is the donor genome.

[0333]

[0334] Table 7: Distribution of related or unrelated genomes P(g 11 , g 21 ). Assumptions 1) All SNPs are from autosomes and 2) All couples are genetically unrelated.

[0335]

[0336] Note that the distributions for parents / children and siblings are very different from unrelated, while the distribution for uncles / aunts / nephews / nieces is almost unrelated. In the case where the donor genotype is unknown, embodiments can infer genetic relationships by evaluating the likelihood function of the fitted model for each of the genetic relationships described above. Alternatively, embodiments can allow for multiple free parameters in the genetic prior distributions (with the additional constraint that the marginal distributions should follow Hardy-Weinberg equilibrium) and estimate these parameters together with the estimate of the donor fraction.

[0337] Integration of modeling components

[0338] The components of the probabilistic mixture model are integrated to provide a solution to the Contributor DNA Quantification (CDQ) problem. The population allele frequency π for each SNP locus can be obtained from public databases such as dbSNP. If the most informative SNP is chosen, i.e., the SNP with π = 0.5, then π = 0.5 can be set for all loci, and as described in the previous section, let P(g 11 , g 21 ) is the prior distribution of genetic correlation.

[0339] At a schematic level, Figure 2B A block diagram illustrating the various components of a probability mixture model 200 is shown. Some components are optional in some embodiments. The probability mixture model 200 includes a binomial distribution 208 for modeling allele counts of sequencing reads. In some embodiments, the probability mixture model also includes a component for modeling donor-recipient (or recipient) relationships using a genetic correlation prior distribution 202. In some embodiments, the probability mixture model also includes a binomial distribution 204 for modeling DNA extraction allele counts. In some embodiments, the probability mixture model 200 also includes a beta distribution 206 for modeling PCR product or amplification product allele scores. See box 206.

[0340] In some embodiments, a mixture model combines binomial distribution 208 with binomial distribution 204 to model both DNA extraction error and sequencing error. In such embodiments, the mixture model uses binomial distribution 210 to model the allele counts of sequencing reads, which allele counts of sequencing reads depend on the allele counts of DNA extraction.

[0341] In some embodiments, the probabilistic mixture model 200 combines the Beta distribution 206 and the Binomial distribution 208 and uses the Beta-Binomial distribution 212 to model both errors in the PCR or amplification process and errors in the sequencing process.

[0342] In some embodiments, the probabilistic mixture model 200 combines the binomial distribution 204, the beta distribution 206, and the binomial distribution 208 to account for variances caused by the DNA extraction, amplification process, and sequencing process, respectively. In such embodiments, the probabilistic mixture model 200 first uses the beta distribution 214 to approximate the effects of the binomial distribution 204 and the beta distribution 206. Then, the probabilistic mixture model 200 combines the beta distribution 214 and the binomial distribution 208 using the beta-binomial distribution 216.

[0343] Sequence Model

[0344] The basic version of the full model ignores the DNA extraction model and the PCR model and only considers the sequencing model. For each locus, the sequencing read counts of the reference allele are given by a binomial distribution ( Figure 1 C)n i1 ~BN(n i1 , p i1 ) model, where the parameter p i1 (g 11 , g 21 , λ, β) is a function of the donor-recipient genotype combination at the locus (Tables 2 and 3). Assuming the genotype is unknown, for each locus, 11 , g 21 |π) as the prior distribution (Table 6 and Table 7), this embodiment marginalizes the 9 possible genotype combinations. The complete likelihood function at all loci is the product of the marginal distributions of all loci:

[0345] L(n1,n2|β,π)=Π i ∑ g11g21 BN(n i1 , p i1 (g 11 , g 21 ,λ,β))·P(g 11 , g 21 |π), where π is a known parameter and β is the donor DNA fraction.

[0346] Extract-Seq Composite Model

[0347] More advanced model combines DNA extraction model and sequencing model.This embodiment ignores PCR step (that is, assuming that for each locus, the allele fraction in the PCR product is identical with the allele fraction in the DNA sample), and only DNA sampling and sequencing step are modeled.For each locus, there is a binomial distribution of allele counts in the input DNA sample.This captures the locus to locus variability of the allele fraction in the input DNA that is provided to NGS sequencing.

[0348] For the DNA extraction model, this embodiment has n i1 ″~BN(n″,p i1 ), and the sequencing model is n i1 |n i1 ″,n i ″~BN(n i , n i1 ″ / n i ″), where n i "=n" is the number of copies of the haploid genome corresponding to the input DNA. Unfortunately, n i1 There is no closed form formula for the marginal distribution of n. This implementation uses the Beta distribution Beta(a, b) to approximate n i1 ″ / n″ distribution, and by changing n i1 The mean and variance of ″ / n″ are similar to those obtained from the binomial model n i1 ″~BN(n″,p i1 ) to select the optimal Beta distribution by matching the mean and variance obtained:

[0349] p i1 =a / (a+b)

[0350] p i1 ·(1-p i1 ) / n″=ab / (a+b) 2 / (a+b+1).

[0351] Solving the equation gives the Beta distribution Beta((n″-1)p i1 , (n″-1)p i2 ) is the best approximation. Approximate to the DNA extraction model, then n i1 The marginal distribution of follows a beta-binomial distribution of the following form:

[0352] BB(n i1 , n i2 |n i , (n″-1)·p1(g 11 , g 21 ,λ,β),(n″-1)·p2(g 11 , g21 , λ, β).

[0353] Then the corresponding total likelihood function considering the genetic correlation prior is:

[0354] L(n1,n2|β,n″,λ,π)=Π i Σ g11g21 BB(n i1 , n i2 |n i , (n″-1)·p1(g 11 , g 21 ,λ,β),(n″-1)·p2(g 11 , g 21 ,λ,β))·P(g 11 , g 21 |π).

[0355] Note that n″ and π=0.5 are both known parameters, and the final full likelihood function has only a single unknown parameter β, the donor DNA fraction.

[0356] The number of copies of the input DNA (haploid) n" can be derived from the mass of the input DNA. When the amount of input DNA is 8 ng, n" = 8 ng / [3.59 x 10 -3 ng / copy]=2228.412.

[0357] PCR-Seq Multiplex Model

[0358] Ignoring the DNA extraction model and assuming a known genotype combination for a given locus, the PCR model is: n i1 ′ / (n i1 ′+n i2 ′)~Beta(n″·ρ i ·p i1 , n″·ρ i ·p i2 ) and sequencing model n i1 ~BN(n i , n1′ / (n1′+n2′)) can be combined into the Beta-binomial distribution BB: (n i1 , n i2 |n i , n″·ρ i ·p i1 , n″·ρ i ·p i2 ). Note that both basic locus-specific PCR amplification rates ρi are unknown. If the embodiment assumes that all loci have the same intrinsic amplification rate, the embodiment will have BB(n i1 , n i2 |ni , c.p i1 (g 11 , g 21 , β), c·p i2 (g 11 , g 21 , β)).

[0359] Then, the full likelihood model over all loci is:

[0360] L(n1,n2|β,c,π)=Π i Σ g11g21 [BB(n i1 , n i2 |n i , c·p1(g 11 , g21,λ,β),c·p2(g 11 ,g21,λ,β))·P(g 11 , g 21 |π)], where c and β are the two parameters to be estimated.

[0361] Alternatively, embodiments may define the relative amplification rate of each locus as proportional to the total reads per locus and reparameterize the beta-binomial as BB(n i1 , n i2 |n i , c′·n i ·p i1 (g 11 , g 21 , β), c′·n i ·p i2 (g 11 , g 21 , β)).

[0362] Then, the full likelihood model over all loci is:

[0363] L(n1,n2|β,c′,π)=Π i ∑ g11g21 [BB(n i1 , n i2 |n i , c′·n i p1(g 11 ,g21,λ,β),c′·n i ·p2(g 11 ,g21,λ,β))·P(g 11 , g 21 |π)], where c and β are the two parameters to be estimated.

[0364] Extraction-PCR-Seq Multiplex Model

[0365] If an embodiment combines the DNA extraction and PCR models into one model and approximates them with a single beta distribution, all three components in the extraction-PCR-sequencing general experimental pipeline can be modeled together by the beta-binomial. Intuitively, although the expected value of the allele 1 fraction in the PCR product (n1′ / n′, see Table 4) is still p1, the uncertainty (variance) of n1′ / n′ comes from both the DNA extraction and PCR steps. In order to obtain a beta distribution beta(a, b) to model DNA extraction and PCR together, the embodiment calculates n based on the following law i1 The unconditional mean and variance of ′ / n′: E(n i1 ' / n')=E(E(n i1 ′ / n i ′|n i1 ′ / n″) and var(n i1 ' / n') = var(E(n i1 ′ / n i ′|n i1 ″ / n″))+E(var(n i1 ′ / n i ′|n i1 ″ / n″)). This gives: E(n i1 ′ / n′)=p i1 , and var(n i1 ′ / n′)=p i1 p i2 / n″+p i1 p i2 / (n″·ρ i +1)-p1p2 / [n″·(n″·ρ i +1)], where ρ i =(1+r i ) / (1-r i )>1 is related to the amplification rate r i Since n″ is large, the implementation has the following approximate value var(n i1 ′ / n′)=p i1 p i2 / [n″·(1+r i ) / 2]. Then, the optimal Beta distribution for simulating DNA extraction and PCR is Beta([n″·(1+r i ) / 2-1]p i1 ,[n″·(1+r i ) / 2-1]p i2 ). Note that this is different from the Beta((n″-1)p i1 , (n″-1)pi2 ) distribution is very close, but the variance is now larger. i = 0.8 to 0.95 for a typical PCR reaction, embodiments have n″·(1+r i ) / 2=0.9·n″ to 0.975·n″.

[0366] The full likelihood function for the cfDNA-PCR-Seq model is:

[0367] L(n1,n2|β,n″,π)=П i ∑ g11g21 BB(n i1 , n i2 |n i ,(n″·(1+r i ) / 2-1)·p1(g 11 , g21, λ, β), (n″·(1+r i ) / 2-1)·p2(g 11 ,g21,λ,β))·P(g 11 , g 21 |π).

[0368] Algorithms for estimating contributor nucleic acid fractions and their confidence intervals

[0369] Numerical optimization for estimating contributor DNA fractions

[0370] The contributor DNA fraction β is estimated as the value that maximizes the total likelihood function L(n1, n2|β). As described above, although DNA is referred to in this and other examples, RNA and other nucleic acid molecules can be similarly processed and analyzed. Similarly, although the examples involve nucleic acid mixture samples, the sample may only include nucleic acid from a single contributor, in which case the contributor fraction will be estimated to be 1 or within an error margin from 1. During the calculation of L(n1, n2|β), multiple small probability values ​​are multiplied. In order to avoid numerical underflow when multiplying small probabilities, this embodiment performs all summations and multiplications on a logarithmic scale. The sum of small probabilities on a logarithmic scale is performed as follows. 1) Obtain the maximum value of the logarithmic probability as x max ; 2) Subtract the maximum value from all log probabilities; 3) Raise the resulting value to a power and then sum; 4) Logarithmically transform the sum; 5) Add back the maximum log probability. log(exp(x1-x max )+exp(x2-x max )+…+exp(x n -x max ))+x max .

[0371] To avoid negative values, use the transformation β = 1 / (1+e-η ), and in order to avoid local minima, use β0 = 1 / (1+e -η 0) Initialize the total likelihood function, where η0 is a value between -10, -9.9, -9.8, ..., -0.1, 0, which makes L(n1, n2|1 / (1+e -η 0)) is maximized. Further numerical optimization of η is performed using the BFGS quasi-Newton method for minimizing -log2(L).

[0372] Estimating confidence intervals

[0373] The lower bound of the confidence interval of the estimate is determined based on the Cramer-Rao inequality: var(θ ML )≥1 / I(θ ML ), where θ ML is the maximum likelihood estimate of the parameter θ, and I(θ ML ) is θ ML Fisher information at . Based on this, the variances of β and c in the above likelihood function can be estimated. Following the Cramér-Rao bound, the standard error is estimated as sqrt(1 / H), where H is the Hessian matrix that can be approximated and estimated using the BFGS-quasi-Newton method.

[0374] We use the following reparameterization to estimate β and c during numerical optimization,

[0375] β=1 / (1+e -η ),

[0376] c=e κ .

[0377] Let I(η) and I(κ) be the Fisher information under parameterization η and κ, then the Fisher information of the original parameters is

[0378] I(β)=I(η)(1 / (β(1-β)) 2

[0379] I(c)=I(k)(1 / c) 2 .

[0380] Therefore, this embodiment performs the following transformation on the estimated standard deviation,

[0381] std(β)=std(η)·β·(1-β)

[0382] std(β)=std(η)·c.

[0383] Deconvoluted Nucleic Acid Samples for Forensic Applications

[0384] Mixture deconvolution: Given observed counts D, infer contributor frequencies f and per-locus genotype configurations G

[0385] The following implementation is suitable for forensic applications. The process described here first obtains the maximum likelihood estimate of f while marginalizing theta and G under their prior conditions (as described below), and then calculates the posterior probability of each genotype configuration conditioned on these estimates. The process reports the MLE of f and the top N lists of reasonable genotype configurations and associated probabilities. The genotype configurations of each locus (over all contributors) and each contributor per locus are reported. The contributor frequency f under this section corresponds to the above-mentioned contributor fraction β. G represents the genotype of the contributor and corresponds to the above-mentioned g 11 , g 21 The data D corresponds to the allele counts n above.

[0386] Sample contains query: Given the observed counts D, the inferred point estimate of f, and the query genotype, infer whether the query genotype is present in the sample.

[0387] Method: The process involves calculating (using known population allele frequencies) the prior probability P(G Q ) and the posterior probability P(G) that the query genotype is in the observed sample (together with other contributors randomly drawn from the population). Q |D), and then report the log ratio between these two probabilities as a measure of evidence. Note that P(G Q |D)=P(D|G Q )P(G Q ) / P(D), so that the reported evidence can also be (equivalently) described as the likelihood ratio P(D|G Q ) / P(D|G random ), because the marginal probability of the data P(D) is the same as that of the data from the population P(D|G random ) have the same probability of being conditional on the data randomly drawn from the contributors.

[0388] Inference Method

[0389] The core computation is a function that takes as input the values ​​of f and theta and computes, for each locus, the marginal log-probability of the data conditioned on f and theta (i.e., the log-likelihood ratio marginalized on G) by performing a sum of genotype-specific probabilities (weighted by the prior genotype). These are then summed over a (discrete equally weighted) distribution of theta to obtain the log-likelihood ratio marginalized on G and theta. The implementation also keeps the sum (over theta) for the individual values ​​of G, so that (after normalization) the implementation has a posterior probability for each genotype at each locus, conditioned on f, but marginalized on theta. The marginal log-likelihoods per locus are accumulated and returned as a single log-likelihood for the entire data set, still conditioned on f. The implementation performs this computation at each point in the grid of possible values ​​of the frequency vector f, and obtains the MLE of f by selecting the grid point where the likelihood is maximal. The frequency grid is set up with equally spaced intervals of 2.5%, plus extra points at all equal frequencies (if not already represented), with the following constraints: the frequencies are listed in non-increasing order and their sum is 1. This produces a 21-point grid for 2 contributors or a 155-point grid for 3 contributors (the 4-contributor case has not yet been implemented and may require a coarser grid; the 3-contributor case can also be accelerated by using a coarser grid first, and then refining as a second step after zooming in on the interesting part of the grid).

[0390] The per-locus posterior probability of the genotype configuration conditioned on the MLE of f is used for the hybrid deconvolution query and the sample query.

[0391] Marginal likelihood calculation:

[0392] The overall log-likelihood is the sum of the locus-specific log-likelihoods: log P(D|theta,f) = \sum l log P(D l |theta, f).

[0393] The locus-specific marginal likelihood is calculated by summing over a large set of plausible genotype configurations: P(D l |theta,f)=\sum G1 P(D l |theta,f,G l )P(G l ). To facilitate the calculation, the following operation is adopted.

[0394] Set the threshold value of the allele that count is less than or equal to 1. In prototype 1, due to historical reasons, this embodiment has constructed both the unpruned list and the pruned list of genotype configuration. Only the unpruned list contains the configuration with allele below the threshold value. Some embodiments only use the pruned list and the shorter version described below (more actively pruned).

[0395] The genotype configurations considered reasonable were limited by constructing a "desired" allele list and listing only those genotype configurations that contained each desired allele at least once. An allele was placed on the desired allele list if its UMI count was judged to be too high to be acceptable via a generic "insert" N-1 jam. The following hard thresholds were used.

[0396] Absolute threshold (count_threshold; set to 10). This is the maximum number of non-stuck UMIs that an embodiment is willing to interpret on alleles that are not present in any contributor.

[0397] Relative threshold (stutter_threshold; set to 0.1). This value multiplied by the count of potential stutter donors (see "Handling Stutters" below) is the maximum number of UMIs that an embodiment is willing to interpret as N-1 stutters.

[0398] If the observed count is above the sum of the two thresholds above, it is considered to be real and must be present in all genotype configurations. If not, it can be omitted (leaving a larger number of genotype configurations to consider).

[0399] Build aggressively pruned lists for use in the more computationally demanding parts (i.e. when inferring contributor frequencies but not for the final deconvolution step):

[0400] For each configuration:

[0401] For each contributor frequency vector, the implementation compares the expected number of reads per allele to the observed number and calculates the RMSE across alleles

[0402] This calculation takes into account lag, but is cheaper than the full likelihood calculation

[0403] Use the minimum RMSE on the frequency vector (is there a reasonably well-formed frequency vector?). For the "optimal RMSE" (see below), embodiments use the average of the frequency vector (which breaks if a best-fit frequency vector is used, which may be completely unrealistic for the entire dataset).

[0404] A configuration is retained if the observations are close enough to the expectations according to the following two criteria:

[0405] Absolute threshold (0.2): prune configurations with RMSE greater than this fraction of the maximum read count

[0406] Relative threshold (5): prune configurations with RMSE greater than this factor from the “best RMSE” ​​(see above)

[0407] The set of plausible genotype configurations and their prior P(G l ) (see below), is constructed during preprocessing and can be reused each time the recognition likelihood calculation is called.

[0408] Genotype-specific likelihood

[0409] The generative model specifies a fixed number of potentially detectable molecules per locus in the "raw" sample (which may correspond to a physical sample collected at a crime scene or later during processing). Embodiments assume that these molecules are divided per allele in proportion to the contributor frequency to which the alleles have been assigned to the contributors in the genotype configuration. Each of these molecules is then detected or not detected so that the number of molecules detected for a given allele is controlled by a binomial process. The probability of detection (i.e., the binomial parameter) may vary between alleles (and between loci), and at each locus, embodiments allocate a beta prior controlled by two parameters:

[0410] Mean (average detection probability)

[0411] Coverage (total number of UMIs detected) varies between loci. In principle, this means that the mean of the beta distribution should vary between loci. Some embodiments keep the mean fixed across loci, but allow the number of molecules in a sample to vary between loci. This should be the same thing (since the two parameters are expected to be highly correlated, treating them as separate parameters would make them almost unidentifiable). The mean parameter is hard-coded to a value of 0.1; the total number of molecules can be set by extrapolating from the observed coverage given this mean parameter and the stutter rate.

[0412] Concentration parameter (how much the probability of detection varies between alleles: this is closely related to, but not the same as, allelic balance as measured in the laboratory)

[0413] This parameter varies between loci and is assigned a 3-component discrete prior (shared across loci) representing low, medium, and high noise conditions.

[0414] Other noise parameters are:

[0415] stutter_prob: For each of the extrapolated total number of molecules (allele N) above, the generative model specifies that it will generate a UMI (allele N-1) with probability stutter_prob.

[0416] This parameter is shared between alleles at the same locus, but varies between loci. It is assigned a 3-component discrete prior (shared across loci) representing low, medium, and high noise conditions.

[0417] expected_dropin: This is the expected number of spurious UMIs observed at an allele (not generated by molecules from that allele or the spurious donor). This parameter is hard-coded (not inferable) and shared across alleles and loci.

[0418] Given the above noise parameters (including theta l ), and then the genotype-specific likelihood P(D l |theta l ,f,G l ) is calculated as the product of allele-specific likelihoods.

[0419] During the likelihood calculation for the locus, some embodiments have tried many genotype configurations, some of which differ from each other only at some alleles. As a result, for the identical counts used before, allele-specific likelihoods are often required. This embodiment stores the result of each allele-specific likelihood calculation in a lookup table, and only calculates the likelihoods whose likelihoods are not in the table.

[0420] We support two types of allele-specific likelihood calculations: a simple (faster) lag-free calculation and a full calculation that takes lags into account.

[0421] Allele-specific likelihood: lag-free version:

[0422] The unstuck version of the likelihood calculation for allele k is applicable if the stuck rate is zero and also used as a computational shortcut when the expected number of "true" (non-stuck) detections is non-zero (due to k being in the genotype of at least one contributor) and the expected number of stuck detections (computed from the stuck rate and the number of molecules assigned to "stuck neighbors" for allele k+1) is below a threshold.

[0423] Ultimately, each molecule assigned to allele k is detected as or not detected as a UMI via a process modeled as a binomial, i.e., detection of individual molecules occurs independently. A binomial process with a frequency parameter of a beta distribution and an allele-specific number of potentially detectable molecules implies a beta-binomial distribution for the observed UMI counts at each allele. Therefore, the allele-specific likelihood is calculated using the formula for the beta-binomial distribution.

[0424] We only compute the probabilities of observed alleles; the penalty caused by observing unobserved alleles with non-zero probability is ignored.

[0425] Dealing with lags and other drop-ins:

[0426] The second likelihood calculation is for the minority of cases determined to be associated with stuck (see above). At allele k with a UMI count of M, some embodiments consider all values ​​m from 0 to M as possible values ​​for the number of UMIs originating from the allele in question, with the remaining (Mm) UMIs originating from allele k+1. The likelihood of one of these cases is the product of the likelihood of the true count and the likelihood of the stuck count (under the binomial model, N equals the number of original UMIs at allele k+1, and the binomial frequency equals the stuck rate). The total likelihood is the (linear domain) sum of all these cases.

[0427] In fact, some embodiments do not need to calculate all the terms in the above sum. This is because the small jam rate makes the distribution of the number of jam observations reach zero quickly (a large number of jam observations is basically impossible, and the corresponding terms in the sum are actually zero). Some embodiments track the cumulative distribution of the number of jam observations and terminate the summation when the remaining probability weight is below a threshold.

[0428] When determining the number of molecules that potentially cause stuttering, general insertions are handled by adding a fixed number of molecules to the number of UMIs assigned to allele k + 1. This number is set to expected_dropin / stutter_prob, so that the expected number of insertion UMIs is equal to expected_dropin.

[0429] Handling missing:

[0430] We distinguish two types of dropout:

[0431] Natural dropout: This is when the binomial process results in an allele count of zero for a contributor's genotype. Rather than explicitly representing every potential allele (which is impossible short of an exhaustive list of potential alleles, or too expensive to do otherwise), some embodiments use a special "dummy" out-of-sample allele. This allele can exist in any genotype configuration, has a UMI count of zero, and is treated as a common allele. Alleles of low-frequency contributors may be naturally dropped out, but alleles of high-frequency contributors are extremely unlikely.

[0432] In order to assign reasonable prior probabilities to out-of-sample alleles, some embodiments make a guess at the total number of potential alleles and set a uniform prior on these potential alleles. Thus, the out-of-sample prior is proportional to the number of unobserved alleles. Currently, a guess at the number of potential alleles is obtained by interpolating all integers between the shortest and longest observed integer-valued alleles, adding any observed non-integer-valued alleles, and returning the maximum of the resulting value and 5.

[0433] Mechanism dropout: Some embodiments incorporate special mechanisms into the model whereby an allele may be "invisible" to the sequencer (e.g., due to a mutation in the primer region), in which case some embodiments do not observe any UMIs regardless of their total number of molecules (i.e., mechanism dropout is just as likely for high frequency contributors as for low frequency contributors). The set of all invisible alleles is represented by a second virtual allele.

[0434] Since this allele cannot be observed, its likelihood is 1 regardless of the data; its only direct contribution to the joint probability is via its prior (which must therefore be set low). For computational convenience, some embodiments use a hard-coded parameter dropout_prob as the prior probability for the invisible allele, and the priors for the visible alleles are calibrated to sum to 1-dropout_prob. This allows the calculation of genotype prior probabilities during preprocessing (see below), with the dropout probability acting as an additional population allele frequency value.

[0435] For monogenic samples, the inferred posterior probability of a homozygous allele depends heavily on the value of dropout_prob, since it determines the probability of the primary alternative hypothesis (a heterozygote with 1 allele) that the model must consider. Based on the intuition that a posterior probability of a monogenic homozygous allele (around 0.999) is reasonable, some implementations calibrate dropout_prob to le-4.

[0436] This allele can exist in any genotypic configuration. It is treated specially when computing the prior for the configuration and is ignored during the likelihood calculation.

[0437] During aggressive pruning of the genotype configuration (see above; based on read counts), the presence of this allele is penalized using a special transformation from its likelihood penalty (depending on dropout_prob) to a "count" value that is intended to be comparable to the mismatch counts estimated for the conventional allele. The transformation is based on a normal approximation to the binomial, by calculating how far from the peak one has to be to incur a likelihood penalty of -log(dropout_prob). The transformation is 1.2876*sqrt(max(read_numbers) / avg_det_prob), where the constant is sqrt(-2p(1-p)ln(D)), where p is avg_det_prob=0.1 and D is dropout-prob=1e-4. Note the dependence on coverage; some embodiments use the coverage of the most abundant allele.

[0438] The list of genotype configurations generated during preprocessing includes genotypes in which one or more alleles are missing alleles. The observed count of a missing allele is always 0, and its likelihood is always 1. Despite the high likelihood, the model does not often use the missing mechanism to explain the data because the low probability of missing results in a low prior.

[0439] Prior and posterior calculations:

[0440] The prior probability of the individual contributor genotype is calculated based on the population allele frequency under the Hardy-Weinberg model: for the homozygous genotype with allele frequency p, P(G) = p^2; for the heterozygous genotype with allele frequencies p and q, P(G) = 2pq. The prior probability of the multi-contributor genotype configuration is the product of each contributor's prior. This is calculated during the preprocessing process when the genotype configuration is constructed and reused in each marginal likelihood calculation. To account for the missing problem, some embodiments add virtual missing alleles with fixed prior probabilities set based on experimental measurements; the population allele frequencies are discounted accordingly.

[0441] The posterior probability of a genotype configuration is the prior multiplied by the likelihood and normalized by the sum over the genotype space (calculated explicitly, since some embodiments calculate each term for each genotype configuration anyway). The posterior probability that a particular contributor (e.g., a major contributor) has a particular genotype is obtained by summing the genotype configurations that contain that genotype.

[0442] For sample inclusion queries, the practitioner is interested in the posterior probability that a specified multiple locus genotype appears in any contributor, provided that it is the same contributor at all loci. This is obtained by summing over all contributors i the probability that contributor i has a specified genotype at all loci (i.e., a ring is a sequence of nested events).

[0443] As used herein, a sample includes "cell-free" (e.g., cfDNA) or cell-bound (e.g., cellular DNA) nucleic acids. Cell-free nucleic acids, including cell-free DNA, can be obtained from biological samples by various methods known in the art, including, but not limited to, plasma, serum, and urine (see, e.g., Fan et al., Proc Natl Acad Sci 105: 16266-16271

[2008] ; Koide et al., Prenatal Diagnosis 25: 604-607

[2005] ; Chen et al., Nature Medicine 2: 1033-1035

[1996] ; Lo et al., Lancet 350: 485-487

[1997] ; Botezatu et al., Clin Chem. 46: 1078-1084, 2000; and Su et al., J Mol. Diagn. 6: 101-107

[2004] ). In order to separate cell-free DNA from cells in a sample, various methods can be used, including but not limited to fractionation, centrifugation (e.g., density gradient centrifugation), DNA-specific precipitation or high-throughput cell sorting and / or other separation methods. Commercially available kits for manual and automatic separation of cfDNA are available (Roche Diagnostics, Indianapolis, IN, Qiagen, Valencia, CA, Macherey-Nagel, Duren, DE). Biological samples containing cfDNA have been used to determine whether there is a chromosomal abnormality, such as trisomy 21, by sequencing assays that can detect chromosomal aneuploidy and / or multiple polymorphisms.

[0444] sample

[0445] As used herein, a sample comprises "cell-free" (e.g., cfDNA) or cell-bound (e.g., cellular DNA) nucleic acids. Cell-free nucleic acids, including cell-free DNA, can be obtained from biological samples by various methods known in the art, including, but not limited to, plasma, serum, and urine (see, e.g., Fan et al., Proc Natl Acad Sci 105: 16266-16271

[2008] ; Koide et al., Prenatal Diagnosis 25: 604-607

[2005] ; Chen et al., Nature Medicine 2: 1033-1035

[1996] ; Lo et al., Lancet 350: 485-487

[1997] ; Botezatu et al., Clin Chem. 46: 1078-1084, 2000; and Su et al., J Mol. Diagn. 6: 101-107

[2004] ). In order to separate cell-free DNA from cells in a sample, various methods can be used, including but not limited to fractionation, centrifugation (e.g., density gradient centrifugation), DNA-specific precipitation or high-throughput cell sorting and / or other separation methods. Commercially available kits for manual and automatic separation of cfDNA are available (Roche Diagnostics, Indianapolis, IN, Qiagen, Valencia, CA, Macherey-Nagel, Duren, DE). Biological samples containing cfDNA have been used to determine whether there is a chromosomal abnormality, such as trisomy 21, by sequencing assays that can detect chromosomal aneuploidy and / or multiple polymorphisms.

[0446] In various embodiments, DNA present in the sample can be enriched specifically or non-specifically before use (e.g., before preparing a sequencing library). Non-specific enrichment of sample DNA refers to whole genome amplification of sample genomic DNA fragments, which can be used to increase the level of sample DNA before preparing a DNA sequencing library. Non-specific enrichment can be the selective enrichment of one of the two genomes present in a sample containing more than one genome. For example, non-specific enrichment can be the selectivity of cancer genomes in plasma samples, which can be obtained by known methods to increase the relative ratio of cancer to normal DNA in the sample. Alternatively, non-specific enrichment can be the non-selective amplification of two genomes present in the sample. For example, non-specific amplification can be the amplification of cancer and normal DNA in a sample, and the sample contains a mixture of DNA from cancer and normal genomes. The method of whole genome amplification is known in the art. Degenerate oligonucleotide-triggered PCR (DOP), primer extension PCR technology (PEP) and multiple displacement amplification (MDA) are examples of whole genome amplification methods. In some embodiments, a sample containing a mixture of cfDNA from different genomes is not enriched for cfDNA of the genome present in the mixture. In other embodiments, a sample comprising a mixture of cfDNA from different genomes is non-specifically enriched for any one genome present in the sample.

[0447] For example, as described above, samples containing nucleic acids to which the methods described herein are applied generally include biological samples ("test samples").

[0448] Thus, in certain embodiments, the sample comprises or consists of: purified or isolated polynucleotides, or it may comprise a sample, such as a tissue sample, a biological fluid sample, a cell sample, etc. Suitable biological fluid samples include, but are not limited to, blood, plasma, serum, sweat, tears, sputum, urine, sputum, ear flow, lymph, saliva, cerebrospinal fluid, wound exudate (ravages), bone marrow suspension, vaginal flow, transcervical lavage, brain fluid, ascites, milk, secretions of the respiratory tract, intestinal and urogenital tracts, amniotic fluid, milk, and leukophoresis. In some embodiments, the sample is a sample that is easily obtained by a non-invasive process, such as blood, plasma, serum, sweat, tears, sputum, urine, sputum, ear flow, saliva, or feces. In certain embodiments, the sample is a peripheral blood sample, or the plasma and / or serum fractions of a peripheral blood sample. In other embodiments, the biological sample is a swab or smear, a biopsy specimen, or a cell culture. In another embodiment, the sample is a mixture of two or more biological samples, for example, the biological sample can include two or more biological fluid samples, tissue samples and cell culture samples. As used herein, the terms "blood", "plasma" and "serum" clearly encompass fractions or processed portions thereof. Similarly, in the case of samples obtained from biopsies, swabs, smears, etc., "sample" clearly encompasses processed fractions or portions obtained from biopsies, swabs, smears, etc.

[0449] In certain embodiments, samples can be obtained from sources including, but not limited to, samples from different individuals, samples from different developmental stages of the same or different individuals, samples from different diseased individuals (e.g., individuals suffering from cancer or suspected of having a genetic disease), samples from normal individuals, samples obtained at different stages of an individual's disease, samples obtained from individuals undergoing different treatments for a disease, samples from individuals exposed to different environmental factors, samples from individuals with susceptibility to a pathology, samples from individuals exposed to an infectious pathogen (e.g., HIV), etc.

[0450] In an exemplary but non-limiting embodiment, the sample is a maternal sample obtained from a pregnant woman such as a pregnant woman. In this case, the method described herein can be used to analyze the sample to provide a prenatal diagnosis of a potential chromosome abnormality of the fetus. The maternal sample can be a tissue sample, a biological fluid sample, or a cell sample. As a non-limiting example, biological fluids include blood, plasma, serum, sweat, tears, sputum, urine, sputum, ear flow, lymph, saliva, cerebrospinal fluid, wound exudate, bone marrow suspension, vaginal flow, secretions of cervical lavage fluid, brain fluid, ascites, milk, respiratory tract, intestinal and urogenital tract, and leukocyte samples.

[0451] In another illustrative but non-restrictive embodiment, the maternal sample is a mixture of two or more biological samples, for example, the biological sample can include two or more biological fluid samples, tissue samples and cell culture samples. In some embodiments, the sample is a sample that is easy to obtain by a non-invasive process, such as blood, plasma, serum, sweat, tears, sputum, urine, milk, sputum, ear flow, saliva and feces. In some embodiments, the biological sample is a peripheral blood sample, and / or its plasma and serum parts. In other embodiments, the biological sample is a swab or smear, a biopsy sample or a cell culture sample. As mentioned above, the terms "blood", "plasma" and "serum" clearly cover its fractions or processed parts. Similarly, in the case of obtaining a sample from a biopsy, a swab, a smear, etc., "sample" clearly covers the processed fractions or parts obtained from a biopsy, a swab, a smear, etc.

[0452] In certain embodiments, samples can also be obtained from tissues, cells or other sources containing polynucleotides cultured in vitro. Cultured samples can be taken from cultures (e.g., tissues or cells) maintained under different culture media and conditions (e.g., pH, pressure or temperature), cultures (e.g., tissues or cells) maintained over different lengths of time, cultures (e.g., tissues or cells) treated with different factors or reagents (e.g., candidate drugs or modulators), or cultures of different types of tissues and / or cells.

[0453] Methods for isolating nucleic acids from biological sources are well known and will vary depending on the nature of the source. As required by the methods described herein, those skilled in the art can easily isolate nucleic acids from sources. In some cases, it may be advantageous to fragment the nucleic acid molecules in the nucleic acid sample. Fragmentation can be random or specific, such as achieved by digestion with restriction endonucleases. Random fragmentation methods are well known in the art and include, for example, limited DNA enzyme digestion, alkali treatment, and physical shearing. In one embodiment, the sample nucleic acid is obtained in the form of cfDNA, which is not fragmented.

[0454] Sequencing library preparation

[0455] In one embodiment, the method described herein can utilize next generation sequencing technology (NGS), which allows multiple samples to be sequenced as genomic molecules (i.e., single sequencing) or a combined sample (e.g., multiple sequencing) comprising indexed genomic molecules in a sequencing run. These methods can generate reads of up to hundreds of millions of DNA sequences. In a plurality of embodiments, the sequence of genomic nucleic acids and / or the genomic nucleic acids of the index can be determined using the next generation sequencing technology (NGS) described herein, for example. In a plurality of embodiments, one or more processors as described herein can be used to perform analysis of a large amount of sequence data obtained using NGS.

[0456] In various embodiments, the use of such sequencing techniques does not involve the preparation of sequencing libraries.

[0457] However, in certain embodiments, the sequencing methods considered herein relate to the preparation of sequencing libraries. In an illustrative method, the preparation of sequencing libraries relates to the production of random collections of adaptor-modified DNA fragments (e.g., polynucleotides) ready to be sequenced. The sequencing libraries of polynucleotides can be prepared from DNA or RNA, including equivalents and analogs of DNA or cDNA, such as DNA or cDNA of complementary or copied DNA produced from RNA templates, by the action of reverse transcriptase. Polynucleotides can originate in double-stranded form (e.g., dsDNA, such as genomic DNA fragments, cDNA, PCR amplification products, etc.), or in certain embodiments, polynucleotides can originate in single-stranded form (e.g., ssDNA, RNA, etc.) and have been converted to dsDNA form. For example, in certain embodiments, single-stranded mRNA molecules can be copied into double-stranded cDNA suitable for preparing sequencing libraries. The precise sequence of a primary polynucleotide molecule is generally unimportant to the library preparation method and can be known or unknown. In one embodiment, the polynucleotide molecule is a DNA molecule. More particularly, in certain embodiments, the polynucleotide molecule represents the entire genetic complement of an organism or substantially the entire genetic complement of an organism, and is a genomic DNA molecule (e.g., cellular DNA, cell-free DNA (cfDNA), etc.), typically including intron sequences and exon sequences (coding sequences), as well as non-coding regulatory sequences, such as promoter and enhancer sequences. In certain embodiments, the primary polynucleotide molecule comprises a human genomic DNA molecule, such as a cfDNA molecule present in the peripheral blood of a pregnant subject.

[0458] The preparation of sequencing libraries for some NGS sequencing platforms is facilitated by using polynucleotides containing a specific range of fragment sizes. The preparation of such libraries generally involves fragmentation of large polynucleotides (e.g., cellular genomic DNA) to obtain polynucleotides of the desired size range.

[0459] Fragmentation can be achieved by any of a variety of methods known to those skilled in the art. For example, fragmentation can be achieved by mechanical means, including but not limited to nebulization, sonication, and hydrodynamic shearing. However, mechanical fragmentation typically cleaves the DNA backbone at CO, PO, and CC bonds, resulting in a heterogeneous mix of blunt ends and 3′- and 5′-overhangs with broken CO, PO, and / or CC bonds (e.g., see Alnemri and Liwack, J Biol. Chem 265:17323-17333

[1990] ; Richards and Boyer, J Mol Biol 11:327-240

[1965] ), which may need to be repaired because they may lack the 5′-phosphate required for subsequent enzymatic reactions, such as the ligation of sequencing adapters required to prepare sequencing DNA.

[0460] In contrast, cfDNA typically exists in fragments of less than about 300 base pairs, and therefore, fragmentation is generally not required for generating sequencing libraries using cfDNA samples.

[0461] Typically, whether polynucleotides are forcibly fragmented (e.g., in vitro fragmentation) or exist naturally as fragments, they are converted to blunt-ended DNA with a 5'-phosphate and a 3'-hydroxyl group. Standard protocols, such as those for sequencing using the Illumina platform described elsewhere herein, instruct the user to end-repair sample DNA, to purify the end-repaired products prior to dA tailing, and to purify the dA-tailed products prior to the adapter ligation step of library preparation.

[0462] The various embodiments of the sequence library preparation methods described herein eliminate the need to perform one or more steps that are typically mandatory by standard protocols to obtain modified DNA products that can be sequenced by NGS. The abbreviated method (ABB method), the 1-step method, and the 2-step method are examples of methods for preparing sequencing libraries, which can be found in patent application 13 / 555,037 filed on July 20, 2012, which is incorporated by reference in its entirety.

[0463] Sequencing methods

[0464] In some embodiments, a prepared sample (eg, a sequencing library) is sequenced as part of a process to deconvolute a mixture of nucleic acids. Any of a variety of sequencing techniques can be used.

[0465] Several sequencing technologies are commercially available, such as the hybridization sequencing platform from Affymetrix (Sunnyvale, California) and the synthesis sequencing platform from 454 Life Sciences (Bradford, Connecticut), Illumina / Solexa (from Hayward, California) and Helicos Biosciences (Cambridge, Massachusetts), and the ligation sequencing platform from Applied Biosystems (Foster City, California). In addition to performing single molecule sequencing using synthesis sequencing from Helicos Biosciences, other single molecule sequencing technologies include, but are not limited to, SMRT from Pacific Biosciences. TM Technology, ION TORRENT TM technologies and nanopore sequencing, for example, developed by Oxford Nanopore Technologies.

[0466] Although the automated Sanger method is considered a "first generation" technique, Sanger sequencing, including automated Sanger sequencing, can also be used in the methods described herein. Other suitable sequencing methods include, but are not limited to, nucleic acid imaging techniques, such as atomic force microscopy (AFM) or transmission electron microscopy (TEM). Illustrative sequencing techniques are described in more detail below.

[0467] In an exemplary but non-limiting embodiment, the method described herein includes using Illumina's synthetic sequencing and reversible terminator-based sequencing chemistry (e.g., as described in Bentley et al., Nature 6: 53-59

[2009] ), to obtain sequence information of nucleic acids in a test sample, such as cfDNA in a maternal sample, cfDNA or cellular DNA of a subject screened for cancer, etc. The template DNA can be genomic DNA, such as cellular DNA or cfDNA. In some embodiments, genomic DNA from isolated cells is used as a template and is fragmented into a length of several hundred base pairs. In other embodiments, cfDNA is used as a template, and fragmentation is not required because cfDNA exists as short fragments. For example, fetal cfDNA circulates in the bloodstream in fragments of about 170 base pairs (bp) in length (Fan et al., Clinical Chemistry 56: 1279-1286

[2010] ), and fragmentation of DNA is not required before sequencing. Circulating tumor DNA exists in short fragments, and its size distribution peaks at about 150-170 bp. Illumina's sequencing technology relies on attaching fragmented genomic DNA to a flat, optically transparent surface to which oligonucleotide anchors are bound. The template DNA is end-repaired to generate 5'-phosphorylated blunt ends, and the polymerase activity of the Klenow fragment is used to add a single A base to the 3' end of the blunt phosphorylated DNA fragment. This addition prepares the DNA fragments for connection to the oligonucleotide adapter, which has a single T base overhang at the 3' end to improve the connection efficiency. The adapter oligonucleotide is complementary to the flow cell anchor oligonucleotide (do not confuse it with the anchor / anchor read sequence in the repeated amplification analysis). Under limiting dilution conditions, the adapter-modified single-stranded template DNA is added to the flow cell and fixed by hybridization with the anchor oligonucleotide. The attached DNA fragments are extended and bridge amplified to create an ultra-high density sequencing flow cell with hundreds of millions of clusters, each cluster containing about 1000 copies of the same template. In one embodiment, the randomly fragmented genomic DNA is amplified using PCR before cluster amplification. Alternatively, a genomic library preparation without amplification (e.g., PCR-free) is used, and only cluster amplification is used to enrich randomly fragmented genomic DNA (Kozarewa et al., Nature Methods 6: 291-295

[2009] ). The templates are sequenced using a powerful four-color DNA synthesis sequencing technology that employs reversible terminators and removable fluorescent dyes. Highly sensitive fluorescence detection can be achieved using laser excitation and total internal reflection optics. Short sequence reads of about tens to hundreds of base pairs are aligned to a reference genome, and a specially developed data analysis pipeline software is used to identify unique mappings of short sequence reads to the reference genome.After the first read is complete, the template can be regenerated in situ to perform a second read from the other end of the fragment. Thus, single-end or double-end sequencing of DNA fragments can be used.

[0468] Multiple embodiments of the present disclosure can use synthetic sequencing that allows paired end sequencing.In some embodiments, the sequencing of the synthesis platform of Illumina involves clustering of fragments.Clustering is the process that each fragment molecule is amplified isothermally.In some embodiments, as the example described here, the fragment has two different adapters attached to the two ends of the fragment, and the adapter allows the fragment to hybridize with two different oligonucleotides on the surface of the flow cell swimming lane.The fragment further includes or is connected to two index sequences at the two ends of the fragment, and the index sequence provides a label to identify different samples in multiple sequencing.In some sequencing platforms, the fragment to be sequenced is also referred to as an insert.

[0469] In some embodiments, the flow cell used for clustering in the Illumina platform is a slide with lanes. Each lane is a glass channel coated with two types of oligonucleotides. The first of the two types of oligonucleotides on the surface makes hybridization possible. The oligonucleotide is complementary to the first adapter on one end of the fragment. The polymerase produces a complementary strand of the hybridized fragment. The double-stranded molecule is denatured, and the original template strand is washed away. The remaining strand is parallel to many other remaining strands and is cloned and amplified by bridging application.

[0470] In bridge amplification, the chains overlap and the second adapter region on the second end of the chain hybridizes to a second type of oligonucleotide on the surface of the flow cell. The polymerase generates a complementary chain, forming a double-stranded bridge molecule. This double-stranded molecule is denatured, resulting in two single-stranded molecules tethered to the flow cell by two different oligonucleotides. The process is then repeated over and over again and occurs simultaneously for millions of clusters, resulting in clonal amplification of all fragments. After bridge amplification, the reverse strand is cut and washed away, leaving only the forward strand. The 3′ end is blocked to prevent unwanted priming.

[0471] After clustering, sequencing begins with extension of the first sequencing primer to generate the first read. In each cycle, fluorescently labeled nucleotides compete for addition to the growing chain. Only one is incorporated based on the sequence of the template. After each nucleotide is added, the cluster is excited by a light source and emits a characteristic fluorescent signal. The number of cycles determines the length of the read. The emission wavelength and signal intensity determine the base calls. For a given cluster, all identical chains are read simultaneously. Hundreds of millions of clusters are sequenced in a massively parallel manner. After the first read is completed, the products of the read are washed away.

[0472] In the next step of the protocol involving two index primers, the Index 1 primer is introduced and hybridized to the Index 1 region on the template. The Index region provides identification of the fragments, which is used to demultiplex the samples during multiplex sequencing. The Index 1 read is generated similarly to the first read. After the Index 1 read is completed, the product of the read is washed away and the 3' end of the strand is deprotected. The template strand is then overlapped and bound to the second oligonucleotide on the flow cell. The Index 2 sequence is read in the same manner as Index 1. The Index 2 read product is then washed out at the completion of this step.

[0473] After reading the two indexes, read 2 is initiated by extending the second flow cell oligonucleotide using a polymerase to form a double-stranded bridge. The double-stranded DNA is denatured and the 3′ end is blocked. The initial positive strand is cut and washed away, leaving the reverse strand. Read 2 begins with the introduction of the read 2 sequencing primer. As with read 1, the sequencing steps are repeated until the desired length is reached. The read 2 product is washed away. This entire process generates millions of reads representing all fragments. Sequences from the merged sample library are separated based on the unique index introduced during sample preparation. For each sample, similar fragment reads for base calls are locally clustered. The forward and reverse reads are paired to create continuous sequences. These continuous sequences are aligned to the reference genome for variant identification.

[0474] The synthetic sequencing examples described above involve paired-end reads, which are used in many embodiments of the disclosed methods. Paired-end sequencing involves two reads at both ends of a fragment. When a pair of reads is mapped to a reference sequence, the base pair distance between the two reads can be determined, which can then be used to determine the length of the fragment from which the read was obtained. In some cases, one of a pair of end reads of a fragment spanning two bands will align to one band, while the other will align it to an adjacent band. This becomes less and less common as the bands become longer or the reads become shorter. Various methods can be used to interpret the band membership of these fragments. For example, they can be omitted when determining the fragment size frequency of the band; they can be counted for two adjacent bands; they can be assigned to the band containing the larger number of base pairs of the two bands; or they can be assigned to both bands with a weight related to the portion of the base pairs in each band.

[0475] Paired end reads can use inserts of different lengths (i.e., different fragment sizes to be sequenced). As the default meaning in the present disclosure, paired end reads are used to refer to reads obtained from various insertion lengths. In some cases, it is used to distinguish short insertion paired end reads from long insertion paired end reads, the latter also referred to as partner pairing reads. In some embodiments involving partner pairing reads, two biotin-linked adapters are first attached to the two ends of a relatively long insert (e.g., several kb). Then, the biotin-linked adapter connects the two ends of the insert to form a circular molecule. Then, a sub-fragment containing a biotin-linked adapter can be obtained by further fragmenting the circular molecule. Then, the sub-fragment including the reverse order of the two ends of the original fragment can be sequenced by the same method as the above-mentioned short insertion paired end sequencing. More detailed information on partner pairing sequencing using the Illumina platform is shown in the online publication at the following URL, which is incorporated by reference in its entirety: res|.|illumina|.|com / documents / products / technotes / technote_nextera_matepair_data_processing. Additional information regarding paired-end sequencing can be found in U.S. Pat. No. 7,601,499 and U.S. Patent Publication No. 2012 / 0,053,063, which are incorporated by reference for materials regarding paired-end sequencing methods and apparatus.

[0476] After sequencing the DNA fragments, the sequence reads of a predetermined length, such as 100bp, are mapped to a known reference genome or compared with it. The reads mapped or compared and their corresponding positions on the reference sequence are also referred to as tags. In one embodiment, the reference genome sequence is the NCB136 / hg18 sequence, which can be obtained on the World Wide Web at genome|.|ucsc|.|edu / cgi-bin / hgGateway? org=Human&db=hg18&hgsid=166260105). Alternatively, the reference genome sequence is GRCh37 / hg19, which can be obtained on the World Wide Web at genetic.ucsc.edu / cgi-bin / hgGateway. Other public sequence information sources include GenBank, dbEST, dbSTS, EMBL (European Molecular Biology Laboratory) and DDBJ (DNA Database of Japan). There are many computer algorithms that can be used to align sequences, including but not limited to BLAST (Altschul et al., 1990), BLITZ (MPsrch) (Sturrock & Collins, 1993), FASTA (Person & Lipman, 1988), BOWTIE (Langmead et al., Genome Biology 10: R25.1-R25.10

[2009] ), or ELAND (Illumina, San Diego, CA, USA). In one embodiment, one end of the clonally expanded copy of the plasma cfDNA molecule is sequenced and processed by bioinformatics alignment analysis on an Illumina Genome Analyzer using the Efficient Large-Scale Alignment of Nucleotide Databases (ELAND) software.

[0477] In one illustrative but non-limiting embodiment, the methods described herein include single molecule sequencing technology using Helicos TrueTrue Molecule Sequencing (tSMS) technology (e.g., as described in Harris TD et al., Science 320: 106-109

[2008] ) to obtain sequence information of nucleic acids in a test sample, such as cfDNA in a maternal sample, cfDNA in a subject being screened for cancer, or cellular DNA, etc. In the tSMS technology, a DNA sample is cut into chains of approximately 100 to 200 nucleotides, and then a polyA sequence is added to the 3′ end of each DNA chain. Each chain is labeled by adding fluorescently labeled adenosine nucleotides. The DNA chains are then hybridized to a flow cell that contains millions of oligo T capture sites fixed to the surface of the flow cell. In some embodiments, the density of the template can be approximately 100 million templates / cm 2The flow cell is then loaded into the instrument (e.g. HeliScope TM Sequencer), and the laser illuminates the surface of the circulation cell to show the position of each template. A CCD camera can map the position of the template on the circulation cell surface. The template fluorescent label is then cut and washed away. The sequencing reaction begins by introducing DNA polymerase and fluorescently labeled nucleotides. Oligo T nucleic acid is used as a primer. The polymerase incorporates the labeled nucleotides into the primer in a template-guided manner. The polymerase and unincorporated nucleotides are removed. By imaging the surface of the circulation cell, templates that have been directed to incorporate fluorescently labeled nucleotides can be identified. After imaging, the cutting step will remove the fluorescent label, and then the process is repeated with other fluorescently labeled nucleotides until the desired read length is achieved. Sequence information is collected in each nucleotide addition step. Whole genome sequencing performed by single molecule sequencing technology excludes or generally avoids PCR-based amplification in the preparation of the sequencing library, and the method allows the sample to be measured directly rather than a copy of the sample.

[0478] In another illustrative but non-limiting embodiment, the method described herein includes using 454 sequencing (Roche) (e.g., as described in Margulies, M. et al., Nature 437: 376-380

[2005] ) to obtain sequence information of nucleic acids in a test sample, such as cfDNA in a maternal test sample, cfDNA in a subject screened for cancer, or cellular DNA, etc. 454 sequencing generally involves two steps. In the first step, the DNA is sheared into fragments of about 300-800 base pairs, and the fragments are then blunt-ended. Oligonucleotide adapters are then attached to the ends of the fragments. The adapters are used as primers for amplification and sequencing of the fragments. The fragments can be attached to DNA capture beads, such as streptavidin-coated beads, using, for example, adapter B containing a 5′-biotin tag. The fragments attached to the beads are PCR amplified in droplets of an oil-water emulsion. The result is that there are multiple copies of the cloned amplified DNA fragments on each bead. In the second step, the beads are captured in a well (e.g., a picoliter-sized well). Pyrophosphate sequencing is performed in parallel on each DNA fragment. The addition of one or more nucleotides generates a light signal that is recorded in the sequencer by a CCD camera. The signal intensity is proportional to the number of nucleotides incorporated. Pyrophosphate sequencing utilizes pyrophosphate (PPi), which is released after the addition of nucleotides. In the presence of 5' adenosine 5'phosphosulfate, PPi is converted to ATP by ATP sulfurylase. Luciferase uses ATP to convert luciferin to oxyluciferin, and the reaction generates light that can be measured and analyzed.

[0479] In another illustrative but non-limiting embodiment, the methods described herein include using SOLiD TMTechnology (Applied Biosystems) to obtain sequence information of nucleic acids in test samples, such as cfDNA in maternal test samples, cfDNA or cellular DNA in subjects screened for cancer, etc. TM In ligation sequencing, genomic DNA is sheared into fragments and adapters are attached to the 5' and 3' ends of the fragments to generate a fragment library. Alternatively, internal adapters can be introduced by connecting adapters to the 5' and 3' ends of the fragments, circularizing the fragments, digesting the circularized fragments to generate internal adapters and attaching adapters to the 5' and 3' ends of the resulting fragments to generate a partner pairing library. Next, a clonal bead population is prepared in a microreactor containing beads, primers, templates, and PCR components. After PCR, the template is denatured and the beads are enriched to separate beads with extended templates. The template on the selected beads is 3' modified, which allows binding to a slide. The sequence can be determined by sequentially hybridizing and connecting partially random oligonucleotides to bases (or base pairs) determined by the center identified by a specific fluorophore. After recording the color, the connected oligonucleotides are cut and removed, and the process is then repeated.

[0480] In another illustrative but non-limiting embodiment, the methods described herein include using Pacific Biosciences' Single Molecule Real-Time (SMRT TM ) sequencing technology, to obtain sequence information of nucleic acids in test samples, such as cfDNA in maternal test samples, cfDNA or cellular DNA in subjects screened for cancer, etc. In SMRT sequencing, the continuous incorporation of dye-labeled nucleotides is imaged during DNA synthesis. Single DNA polymerase molecules are attached to the bottom surface of each zero-mode wavelength detector (ZMW detector), which obtains sequence information when phosphorylated nucleotides are incorporated into the growing primer chain. The ZMW detector includes a restriction structure that enables the incorporation of single nucleotides by DNA polymerase against the background observation of fluorescent nucleotides that diffuse rapidly in and out of the ZMW (e.g., in microseconds). It usually takes several milliseconds to incorporate nucleotides into the growing chain. During this period, the fluorescent label is excited and produces a fluorescent signal, and the fluorescent label is cut off. The measurement of the corresponding fluorescence of the dye indicates which base is incorporated. Repeat this process to provide a sequence.

[0481] In another illustrative but non-limiting embodiment, the method described herein includes using nanopore sequencing (e.g., as described in Soni GV and Meller A. Clinical Chemistry 53: 1996-2001

[2007] ) to obtain sequence information of nucleic acids in a test sample, such as cfDNA in a maternal test sample, cfDNA in a subject screened for cancer, or cellular DNA, etc. Nanopore sequencing DNA analysis technology is developed by many companies, including, for example, Oxford Nanopore Technologies (Oxford, UK), Sequenom, NABsys, etc. Nanopore sequencing is a single-molecule sequencing technology by which single-molecule DNA can be directly sequenced as it passes through a nanopore. A nanopore is a small hole, typically about 1 nanometer in diameter. The nanopore is immersed in a conductive fluid and an electric potential (voltage) is applied thereto, resulting in a small amount of current being generated due to the conduction of ions through the nanopore. The amount of current flowing is sensitive to the size and shape of the nanopore. When a DNA molecule passes through a nanopore, each nucleotide on the DNA molecule blocks the nanopore to varying degrees, thereby changing the magnitude of the current passing through the nanopore to varying degrees. Therefore, this change in the current as the DNA molecule passes through the nanopore provides a read of the DNA sequence.

[0482] In another illustrative but non-limiting embodiment, the method described herein includes using a chemically sensitive field effect transistor (chemFET) array (e.g., as described in U.S. Patent Application Publication No. 2009 / 0026082) to obtain sequence information of nucleic acids in a test sample, such as cfDNA in a maternal test sample, cfDNA in a subject being screened for cancer, or cellular DNA, etc. In one example of this technology, a DNA molecule can be placed in a reaction chamber, and the template molecule can be hybridized with a sequencing primer bound to a polymerase. The incorporation of one or more triphosphates into a new nucleic acid chain at the 3′ end of the sequencing primer can be recognized by a chemFET as a change in current. The array can have multiple chemFET sensors. In another example, a single nucleic acid can be attached to a bead, and the nucleic acid can be amplified on the bead, and each bead can be transferred to each reaction chamber on the chemFET array, each chamber having a chemFET sensor, and the nucleic acid can be sequenced.

[0483] In another embodiment, the method includes using transmission electron microscopy (TEM) to obtain nucleic acids in a test sample, such as sequence information of cfDNA in a maternal test sample. The method, known as individual molecule placement rapid nanotransfer (IMPRNT), includes single-atom resolution transmission electron microscopy imaging of high molecular weight (150kb or greater) DNA selectively labeled with heavy atom markers, and arranging these molecules on an ultra-thin film in an ultra-dense (3nm chain to chain) parallel array with consistent base spacing. Electron microscopes are used to image molecules on film to determine the position of heavy atom markers and to extract base sequence information from DNA. The method is further described in PCT patent publication WO 2009 / 046445. The method allows sequencing of a complete human genome in less than ten minutes.

[0484] In another embodiment, the DNA sequencing technology is ion torrent single molecule sequencing, which combines semiconductor technology with simple sequencing chemistry to translate chemically encoded information (A, C, G, T) directly into digital information (0, 1) on a semiconductor chip. In fact, when nucleotides are incorporated into the DNA chain by polymerase, hydrogen ions are released as byproducts. Ion torrent uses a high-density array of micro-machined holes to perform this biochemical process in a massively parallel manner. There is a different DNA molecule in each hole. Below the hole is an ion-sensitive layer, and below the ion-sensitive layer is an ion sensor. When a nucleotide (such as C) is added to the DNA template and then incorporated into the DNA chain, a hydrogen ion will be released. The charge of this ion will change the pH of the solution, which can be detected by the ion sensor of ion torrent. The sequencer is essentially the world's smallest solid-state pH meter - the base is detected, directly converted from chemical information to digital information. Then, the ion personal genome machine (PGM) TM ) The sequencer floods the chip with one nucleotide after another. If the next nucleotide flooding the chip does not match, no voltage change will be recorded and no base will be detected. If there are two identical bases on the DNA strand, the voltage will double and the chip will record two identical bases detected. Direct detection allows nucleotide incorporation to be recorded in seconds.

[0485] In another embodiment, the method includes obtaining nucleic acids in a test sample by hybridization sequencing, such as sequence information of cfDNA in a maternal test sample. Hybridization sequencing includes contacting a plurality of polynucleotide sequences with a plurality of polynucleotide probes, wherein each of the plurality of polynucleotide probes can be optionally bound to a substrate. The substrate may be a flat surface comprising an array of known nucleotide sequences. The hybridization pattern of the array can be used to determine the polynucleotide sequences present in the sample. In other embodiments, each probe is bound to a bead, such as a magnetic bead, etc. Hybridization with the beads can be determined and used to identify a plurality of polynucleotide sequences in the sample.

[0486] In some embodiments of the method described herein, the sequence tag mapped comprises about 20bp, about 25bp, about 30bp, about 35bp, about 40bp, about 45bp, about 50bp, about 55bp, about 60bp, about 65bp, about 70bp, about 75bp, about 80bp, about 85bp, about 90bp, about 95bp, about 100bp, about 110bp, about 120bp, about 130, about 140bp, about 150bp, about 200bp, about 250bp, about 300bp, about 350bp, about 400bp, about 450bp, or about 500bp sequencing reads. It is expected that technological progress will make the single-end read greater than 500bp possible, thereby allowing the read greater than about 1000bp when generating paired end reads. In one embodiment, the sequence tag mapped comprises the sequence read of 36bp. Sequence tag mapping is achieved by comparing the sequence of the tag with the reference sequence to determine the chromosomal origin of the sequenced nucleic acid (e.g., cfDNA) molecule, and no specific genetic sequence information is required. A small degree of mismatch (0-2 mismatches per sequence tag) can be allowed to account for minor polymorphisms that may exist between the reference genome and the genome in the mixed sample.

[0487] Typically, multiple sequence tags are obtained per sample. In some embodiments, at least about 3×10 6 sequence tags, at least about 5×10 6 sequence tags, at least about 8×10 6 sequence tags, at least about 10×10 6 sequence tags, at least about 15×10 6 sequence tags, at least about 20×10 6 sequence tags, at least about 30×10 6 sequence tags, at least about 40×10 6 sequence tags or at least about 50 × 10 6In one embodiment, all sequence reads are mapped to all regions of the reference genome. In one embodiment, labels mapped to all regions of the reference genome (e.g., all chromosomes) are analyzed.

[0488] Device and system for deconvoluting mixtures of nucleic acids from multiple sources

[0489] The analysis of sequencing data and the diagnosis derived therefrom are usually performed using algorithms and programs executed by various computers. Therefore, some embodiments adopt the processing of data stored in one or more computer systems or other processing systems or transmitted therethrough. Embodiments disclosed herein also relate to devices for performing these operations. The device can be specially constructed for the desired purpose, or it can be a general-purpose computer (or a group of computers) selectively activated or reconfigured by a computer program and / or data structure stored in a computer. In some embodiments, a group of processors performs some or all of the listed analysis operations collaboratively (e.g., via a network or cloud computing) and / or in parallel. A processor or a group of processors for performing the method described herein can be of various types, including microcontrollers and microprocessors, such as programmable devices (e.g., CPLD and FPGA) and non-programmable devices, such as gate array ASICs or general-purpose microprocessors.

[0490] In addition, some embodiments relate to tangible and / or non-transient computer-readable media or computer program products, which include program instructions and / or data (including data structures) for performing various computer-implemented operations. Examples of computer-readable media include, but are not limited to, semiconductor memory devices, magnetic media such as disk drives, tapes, optical media such as CDs, magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as read-only memory devices (ROM) and random access memory (RAM). Computer-readable media can be directly controlled by end users, or media can be indirectly controlled by end users. Examples of directly controlled media include media located at user facilities and / or media that are not shared with other entities. Examples of indirectly controlled media include media that users can access indirectly via external networks and / or via services that provide shared resources (such as "clouds"). Examples of program instructions include machine code, such as machine code generated by a compiler, and files containing higher-level code that can be executed by a computer using an interpreter.

[0491] In various embodiments, data or information used in the disclosed methods and apparatus are provided in an electronic format. Such data or information may include reads and labels derived from nucleic acid samples, counts or densities of such labels aligned to specific regions of reference sequences (e.g., aligned to chromosomes or chromosome fragments), reference sequences (including reference sequences that provide only or primarily polymorphisms), chromosome and segment doses, detections such as SNV or aneuploidy detections, normalized chromosome and segment values, paired chromosomes or segments and corresponding normalized chromosomes or segments, counseling advice, diagnosis, etc. As used herein, data or other information provided in an electronic format may be used for storage on a machine and transmission between machines. Typically, data in an electronic format is provided in digital form and may be stored as bits and / or bytes in various data structures, lists, databases, etc. The data may be embodied electronically, optically, or the like.

[0492] One embodiment provides a kind of computer program product, for generating the output indicating the presence or absence of SNV or aneuploidy associated with cancer in the test sample.The computer product may include instructions for executing any one or more of the above-mentioned methods for determining chromosome abnormality.As explained, the computer product may include non-transient and / or tangible computer-readable medium with computer executable or compilable logic (e.g., instruction) recorded thereon, so that the processor can deconvolute nucleic acid mixture.In one example, the computer product includes a computer-readable medium, and the computer-readable medium has computer executable or compilable logic (e.g., instruction) recorded thereon, so that the processor can deconvolute nucleic acid mixture.

[0493] The sequence information from the sample under consideration can be mapped to a chromosome reference sequence to identify multiple sequence tags for each chromosome in any one or more target chromosomes, and identify multiple sequence tags for normalizing the segment sequence of each chromosome in any one or more target chromosomes. In multiple embodiments, the reference sequence is stored in a database such as a relational database or an object database.

[0494] It should be understood that it is impractical, or even impossible in most cases, for an unassisted person to perform the computational operations of the methods disclosed herein. For example, mapping a single 30 bp read in a sample to any human chromosome may take years of effort without the aid of a computing device.

[0495] The method disclosed herein can be performed using a system for quantifying a nucleic acid sample containing nucleic acids of one or more contributors. The system includes: (a) a sequencer for receiving nucleic acids from a test sample and providing nucleic acid sequence information from the sample; (b) a processor; and (c) one or more computer-readable storage media having stored thereon instructions for execution on the processor to perform a method for deconvoluting a nucleic acid mixture.

[0496] In some embodiments, the method is indicated by a computer-readable medium having computer-readable instructions for performing a method for deconvoluting a mixture of nucleic acids stored thereon. Thus, one embodiment provides a computer program product comprising one or more computer-readable non-transitory storage media having computer-executable instructions stored thereon, which, when executed by one or more processors of a computer system, cause the computer system to implement a method for the following purpose: quantifying a nucleic acid sample comprising one or more contributor nucleic acids. The method comprises: (a) receiving, by a computer system, a nucleic acid sequence read of one or more alleles obtained from a nucleic acid sample and mapped to one or more polymorphic loci; (b) using the nucleic acid sequence read and determining, by one or more processors, an allele count of each of the one or more alleles at the one or more polymorphic loci; and (c) quantifying one or more fractions of nucleic acids of one or more contributors in a nucleic acid sample using a probabilistic mixture model and by one or more processors, wherein using the probabilistic mixture model comprises applying the probabilistic mixture model to the allele counts, and the probabilistic mixture model models the allele counts at the one or more polymorphic loci using a probability distribution that takes into account errors in the nucleic acid sequence read.

[0497] In some embodiments, the instruction may further include automatically recording information related to the method in the patient's medical record for the human subject providing the maternal test sample. The patient's medical record may be maintained by, for example, a laboratory, a physician's office, a hospital, a health maintenance organization, an insurance company, or a personal medical record website. In addition, based on the result of the analysis implemented by the processor, the method may further include prescribing, starting, and / or changing treatment of the human subject from which the maternal test sample has been obtained. This may involve carrying out one or more other tests or analyses on other samples collected from the subject.

[0498] The disclosed method can also be performed using a computer processing system that is suitable for or configured to perform a method for quantifying a nucleic acid sample comprising nucleic acids of one or more contributors. One embodiment provides a computer processing system that is suitable for or configured to perform the method described herein. In one embodiment, the device includes a sequencing device that is suitable for or configured to sequence at least a portion of nucleic acid molecules in a sample to obtain the type of sequence information described elsewhere herein. The device may also include a component for processing a sample. Such components are described elsewhere herein.

[0499] Sequence or other data can be directly or indirectly input to a computer or stored on a computer readable medium. In one embodiment, a computer system is directly coupled to a sequencing device, and this sequencing device reads and / or analyzes the nucleotide sequence from a sample. The sequence or other information from this type of instrument is provided via the interface in the computer system. Alternatively, the sequence of system processing is provided from a sequence storage source (such as a database or other storage repository). Once it can be used for a processing device, a memory device or a large-capacity storage device at least temporarily buffers or stores the nucleotide sequence. In addition, a memory device can store the label counts of various chromosomes or genomes, etc. Memory can also store various routines and / or programs for analyzing and presenting sequences or mapping data. This type of program / routine can include programs for performing statistical analysis, etc.

[0500] In one example, a user provides a sample to a sequencing device. Data is collected and / or analyzed by a sequencing device connected to a computer. Software on the computer allows data collection and / or analysis. Data can be stored, displayed (via a monitor or other similar device) and / or sent to another location. The computer can be connected to the Internet, which is used to transmit data to a handheld device used by a remote user (e.g., a physician, scientist, or analyst). It should be understood that data can be stored and / or analyzed before transmission. In some embodiments, raw data is collected and sent to a remote user or device that will analyze and / or store data. Transmission can be performed via the Internet, or via satellite or other connections. Alternatively, data can be stored on a computer-readable medium, and the medium can be shipped to an end user (e.g., via mail). Remote users can be in the same or different geographical locations, including but not limited to buildings, cities, states, countries, or continents.

[0501] In some embodiments, the method also includes collecting data about multiple polynucleotide sequences (for example, read order, label and / or reference chromosome sequence), and sending the data to a computer or other computing system. For example, a computer can be connected to laboratory equipment, such as a sample collection device, a nucleotide amplification device, a nucleotide sequencing device or a hybridization device. Then, the computer can collect the applicable data collected by the laboratory equipment. Data can be stored on a computer at any step, for example, before sending, during sending, or in conjunction with real-time information collection during or after sending. Data can be stored on a computer-readable medium that can be extracted from a computer. The collected or stored data can be, for example, transferred to a remote location from a computer via a local area network or a wide area network such as the Internet. As described below, various operations can be performed on the data transmitted at a remote location.

[0502] The types of electronically formatted data that may be stored, transmitted, analyzed, and / or manipulated in the systems, apparatus, and methods disclosed herein are as follows:

[0503] The reads obtained by sequencing the nucleic acids in the test sample

[0504] Tags obtained by aligning reads to a reference genome or one or more other reference sequences

[0505] Reference genome or sequence

[0506] Sequence tag density - the count or number of tags in each of two or more regions (usually chromosomes or chromosome segments) of a reference genome or other reference sequence.

[0507] The identity of the chromosome or chromosome segment normalized to the specific chromosome or chromosome segment of interest

[0508] Doses of chromosomes or chromosome segments (or other regions) obtained from the target chromosome or segment and the corresponding normalizing chromosome or segment

[0509] Thresholds for calling chromosome dosage as affected, unaffected, or no call

[0510] Actual determination of chromosome dosage

[0511] Diagnosis (clinical condition associated with detection)

[0512] Recommendations for further testing resulting from detection and / or diagnosis

[0513] Treatment and / or surveillance plan derived from detection and / or diagnosis

[0514] These various types of data can be obtained, stored, transmitted, analyzed, and / or manipulated at one or more locations using different devices. The processing options span a wide spectrum. At one end of the spectrum, all or much of this information is stored and used at the location where the test sample is processed, such as a doctor's office or other clinical location. At the other extreme, the sample is obtained at one location, can be processed and optionally sequenced at a different location, read alignment and detection are performed at one or more different locations, and a diagnosis, recommendation, and / or plan is prepared at another location (which may be the location where the sample was obtained).

[0515] In a number of embodiments, a read is generated with a sequencing device and then transmitted to a remote site where it is processed to generate a call. For example, at the remote location, the read is compared with a reference sequence to generate a label, which is counted and assigned to a target chromosome or fragment. Also at the remote location, the count is converted to a dose using an associated normalized chromosome or fragment. Further, at the remote location, the dose is used to generate a call.

[0516] The processing operations that can be adopted at different locations are as follows:

[0517] Sample collection

[0518] Sample processing before sequencing

[0519] Sequencing

[0520] Analyze sequence data and quantify nucleic acid samples containing one or more contributor nucleic acids

[0521] diagnosis

[0522] Report the diagnosis and / or detection to the patient or healthcare provider

[0523] Develop a plan for further treatment, testing, and / or monitoring

[0524] Execution plan

[0525] consult

[0526] Any one or more of these operations can be automatically performed as described in this other place. Usually, the order-checking and analysis of the DNA mixture sample of the sequence data and deconvolution are performed by calculation. Other operations can be performed manually or automatically.

[0527] Examples of locations where sample collection can be performed include medical practitioners' offices, clinics, patients' homes (places where sample collection tools or test kits are provided), and mobile medical vehicles. Examples of locations where sample processing can be performed before sequencing include medical practitioners' offices, clinics, patients' homes (places where sample processing devices or test kits are provided), mobile medical vehicles, and facilities of DNA analysis providers. Examples of locations where sequencing can be performed include medical practitioners' offices, clinics, medical practitioners' offices, clinics, patients' homes (places where sample sequencing devices and / or test kits are provided), mobile medical vehicles, and facilities of DNA analysis providers. A dedicated network connection can be provided at the location where sequencing is performed for transmitting sequence data (usually read sequences) in an electronic format. Such connections can be wired or wireless, and have and can be configured to send data to sites that can process and / or aggregate data before being transmitted to a processing site. Data aggregators can be maintained by health organizations such as health maintenance organizations (HMOs).

[0528] The analysis and / or derivation operation can be performed at any one of the aforementioned locations, or alternatively at another remote site dedicated to the service of calculating and / or analyzing nucleic acid sequence data. Such locations include, for example, facilities such as clusters of general server farms, DNA analysis service businesses, etc. In some embodiments, the computing device for performing the analysis is leased or rented. Computing resources can be part of the Internet accessible set of processors, such as the processing resources commonly known as clouds. In some cases, the calculation is performed in parallel or on a large scale by a group of mutually related or unrelated processors. Distributed processing such as cluster computing, grid computing, etc. can be used to complete the processing. In such embodiments, a cluster or grid of computing resources jointly forms a super virtual computer composed of multiple processors or computers, and the multiple processors or computers act together to perform analysis and / or derivation described herein. As described herein, these technologies and more conventional supercomputers can be used to process sequence data. Each is a parallel computing form that relies on a processor or a computer. In the case of grid computing, these processors (usually entire computers) are connected by a network (dedicated, public or Internet) through a conventional network protocol (such as Ethernet). In contrast, a supercomputer has many processors connected by a local high-speed computer bus.

[0529] In certain embodiments, the diagnosis is generated at the same location as the analytical operation. In other embodiments, it is performed at different locations. In certain examples, the report diagnosis is performed at the location of sampling, although not necessarily so. The example of the location that can generate or report diagnosis and / or execute the plan includes the office of medical practitioners, clinics, computer-accessible Internet sites and handheld devices, such as mobile phones, tablet computers, smart phones, etc. with wired or wireless connections of the network. The example of the location of the execution consultation includes the office of health practitioners, clinics, computer-accessible Internet sites, handheld devices, etc.

[0530] In some embodiments, sample collection, sample processing, and sequencing operations are performed at a first location, and analysis and derivation operations are performed at a second location. However, in some cases, sample collection is collected at one location (e.g., a health practitioner's office or clinic), and sample processing and sequencing are performed at a different location, which is optionally the same location where analysis and derivation are performed.

[0531] In a number of embodiments, the order of operations listed above can be triggered by starting sample collection, sample processing and / or sequencing by a user or entity. After one or more of these operations begin to execute, other operations will naturally follow. For example, a sequencing operation may cause the read sequence to be automatically collected and sent to a processing device, which then generally automatically and possibly performs sequence analysis without further user intervention. In some embodiments, the result of the processing operation may then be automatically passed to a system component or entity that reports information to a health professional and / or patient as a diagnosis in a reformatted manner. As explained, such information may also be automatically processed to generate a treatment, test and / or monitoring plan, possibly together with consulting information. Therefore, starting early operations can trigger an end-to-end sequence in which diagnosis, planning, consulting and / or other information useful for a physical condition is provided to a health professional, a patient or other interested party. This can be achieved even if the parts of the entire system are physically separated and may be far away from the location of, for example, a sample and a sequencing device.

[0532] Figure 4A typical computer system according to certain embodiments is shown in a simple box format, which can be used as a computing device when the computer system is appropriately configured or designed. Computer system 2000 includes any number of processors 2002 (also referred to as central processing units or CPUs) coupled to storage devices, which include main storage devices 2006 (usually random access memory or RAM), main storage devices 2004 (usually read-only memory or ROM). CPU 2002 can be of various types, including microcontrollers and microprocessors such as programmable devices (e.g., CPLD and FPGA), and non-programmable devices such as gate array ASICs or general-purpose microprocessors. In the depicted embodiment, main storage device 2004 is used to transmit data and instructions to CPU unidirectionally, and main storage device 2006 is generally used to transmit data and instructions in a bidirectional manner. Both main storage devices can include any suitable computer-readable medium, such as the above-mentioned medium. Mass storage device 2008 is also bidirectionally coupled to main storage device 2006, and provides additional data storage capacity, and can include any of the above-mentioned computer-readable media. The mass storage device 2008 can be used to store programs, data, etc., and is typically an auxiliary storage medium such as a hard disk. Typically, such programs, data, etc. are temporarily copied to the main memory 2006 for execution on the CPU 2002. It will be understood that the information retained in the mass storage device 2008 can be combined as part of the main storage device 2004 in a standard manner where appropriate. Certain mass storage devices such as CD-ROM 2014 can also pass data unidirectionally to the CPU or the main storage device.

[0533] CPU 2002 is also coupled to interface 2010, which is connected to one or more input / output devices, such as nucleic acid sequencer (2020), video monitor, tracking ball, mouse, keyboard, microphone, touch-sensitive display, transducer card reader, magnetic tape or paper tape reader, tablet computer, stylus, voice or handwriting recognition peripheral device, USB port or other well-known input device, such as other computers. Finally, CPU 2002 can optionally be connected to external devices, such as databases, computers or telecommunications networks using external connections (as shown in 2012 places in general). Through this type of connection, it can be expected that CPU may receive information from the network, or may output information to the network during the execution of the method steps described herein. In some embodiments, instead of or in addition to via interface 2010, nucleic acid sequencer (2020) can also be linked to CPU 2002 via network connection 2012 communication ground.

[0534] In one embodiment, a system such as computer system 2000 is used as a data import, data association, and query system capable of performing some or all of the tasks described herein. Information and programs including data files may be provided via network connection 2012 for researchers to access or download. Alternatively, such information, programs, and files may be provided to researchers on a storage device.

[0535] In a specific embodiment, the computer system 2000 is directly coupled to a data acquisition system, such as a microarray, a high throughput screening system, or a nucleic acid sequencer (2020) that captures data from a sample. Data from such a system is provided via an interface 2010 for analysis by the system 2000. Alternatively, data processed by the system 2000 is provided from a data storage source such as a database or other repository of relevant data. Once in the device 2000, a memory device such as a main storage device 2006 or a mass storage device 2008 buffers or stores the relevant data at least temporarily. The memory may also store various routines and / or programs for importing, analyzing, and presenting data, including sequence reads, UMIs, codes for determining sequence reads, overlapping sequence reads, and correcting errors in reads, etc.

[0536] In some embodiments, the computer used herein may include a user terminal, which may be any type of computer (e.g., desktop computer, laptop computer, tablet computer, etc.), media computing platform (e.g., cable, satellite set-top box, digital video recorder, etc.), handheld computing device (e.g., PDA, email client, etc.), mobile phone, or any other type of computing or communication platform.

[0537] In certain embodiments, the computer used here can also include a server system communicated with a user terminal, and the server system can include a server device or a decentralized server device, and can include a mainframe computer, a minicomputer, a supercomputer, a personal computer, or a combination thereof. Without departing from the scope of the present invention, multiple server systems can also be used. The user terminal and the server system can communicate with each other through a network. The network can include, for example, a wired network, such as a LAN (local area network), a WAN (wide area network), a MAN (metropolitan area network), an ISDN (integrated services digital network), etc., and a wireless network, such as a wireless LAN, a CDMA, a Bluetooth and a satellite communication network, etc., without limiting the scope of the present invention.

[0538] Figure 5One embodiment of a decentralized system for generating a detection or diagnosis from a test sample is shown. Sample collection location 01 is used to obtain a test sample. The sample is then provided to a processing and sequencing location 03, where the test sample can be processed and sequenced as described above. Location 03 includes a device for processing the sample and a device for sequencing the processed sample. As described elsewhere herein, the result of sequencing is a collection of reads, which are typically provided in an electronic format and provided to a network such as the Internet, which is available at Figure 5 Indicated by reference numeral 05.

[0539] The sequence data is provided to a remote location 07 where analysis and call generation is performed. This location may include one or more powerful computing devices, such as computers or processors. After the computing resources at location 07 complete the analysis and generate a call based on the received sequence information, the call is relayed back to the network 05. In some embodiments, not only is a call generated at location 07, but an associated diagnosis is also generated. The call and / or diagnosis is then transmitted over the network and returned to the sample collection location 01, such as Figure 5 As explained, this is just one of many variations on how the various operations associated with generating a call or diagnosis are divided between various locations. One common variation involves providing sample collection, processing, and sequencing in a single location. Another variation involves providing processing and sequencing at the same location as analysis and call generation.

[0540] Figure 6 Details the options for performing various operations at different locations. Figure 6 In the most granular sense shown in , each of the following operations is performed at a separate location: sample collection, sample processing, sequencing, read alignment, calling, diagnosis, and reporting and / or planning.

[0541] In one embodiment that aggregates some of these operations, sample processing and sequencing are performed in one location, and read alignment, calling, and diagnostics are performed at separate locations. Figure 6 The portion identified by reference character A. Figure 6 In another embodiment identified by character B in FIG, sample collection, sample processing, and sequencing are all performed at the same location. In this embodiment, read alignment and detection are performed at a second location. Finally, diagnosis and reporting and / or planning are performed at a third location. Figure 6 In the embodiment depicted by letter C in FIG, sample collection is performed at a first location, sample processing, sequencing, read alignment, detection, and diagnosis are all performed together at a second location, and reporting and / or planning is performed at a third location. Figure 6In the embodiment marked D, sample collection is performed at a first location, sample processing, sequencing, read alignment and detection are all performed at a second location, and diagnosis and reporting and / or plan management are performed at a third location.

[0542] One embodiment provides a system for analyzing simple nucleotide variants associated with tumors in cell-free DNA (cfDNA), the system comprising a sequencer for receiving a nucleic acid sample and providing nucleic acid sequence information from the nucleic acid sample; a processor; and a machine-readable storage medium, comprising instructions for execution on the processor, the instructions comprising: (a) a code for receiving a nucleic acid sequence read obtained from the nucleic acid sample and mapped to one or more alleles at one or more polymorphic loci; (b) a code for determining the allele counts of each allele of one or more alleles at one or more polymorphic loci using the nucleic acid sequence reads; and (c) using a probabilistic mixture model encoding to quantify one or more fractions of nucleic acids of one or more contributors in the nucleic acid sample. In some embodiments, using a probabilistic mixture model includes applying a probabilistic mixture model to allele counts. The probabilistic mixture model models the allele counts at one or more polymorphic loci using a probability distribution that takes into account errors in the nucleic acid sequence reads.

[0543] In some embodiments of any system provided herein, the sequencer is configured to perform next generation sequencing (NGS). In some embodiments, the sequencer is configured to perform large-scale parallel sequencing using synthetic sequencing by reversible dye terminators. In other embodiments, the sequencer is configured to perform connection sequencing. In other embodiments, the sequencer is configured to perform single molecule sequencing.

[0544] Experimental

[0545] Example 1

[0546] This example uses data obtained from actual DNA mixture samples to illustrate that some embodiments can provide higher accuracy and reliability, as well as lower empirical bias, when quantifying DNA mixture samples compared to conventional techniques that do not use the probabilistic methods disclosed herein.

[0547] DNA mixture samples included two DNAs from the genome (contributor), and the minor fractions in different samples were 0.1%, 0.2%, 0.4%, and 2%. Some samples contained 3ng of input DNA, and other samples contained 10ng. Samples were processed with two experimental procedures labeled Nack or Nack2 to indicate two primer designs with different numbers of target loci for the two designs. Some samples were processed using the MiSeq sequencing platform, and others were processed using the MiniSeq platform.

[0548] Sample data were analyzed using three different methods. Table 2 shows the average value of the coefficient of variation (CV, defined as standard_deviation_of_predictions / true_fraction) values ​​of multiple mixture fractions using three different methods of various samples and experimental processes, as well as the average value of the coefficient of variation + deviation values ​​of multiple mixture fractions (CVB, usually expressed as CV (RMSD) and defined as RMSD / true_fraction). The first method applies a probability model that includes a binomial distribution for modeling sequence errors. The first method corresponds to some embodiments described above as the Seq model. The data of the first method (Seq) are shown in the third row of Table 8. The second method applies a probabilistic mixture model, including probability distributions that explain DNA extraction errors, PCR amplification errors, and sequencing errors. The second method corresponds to some embodiments described above as the "Extraction-PCR-Seq Model". The fourth row of Table 8 shows the data of the second method (EPS).

[0549] The third method applied a deterministic linear regression model to describe the allele description data. The sum of squared errors of the estimated data was as follows.

[0550] E=[r i -p i (β)]T·[r i -p i (β)]

[0551] where r is the observed allele fraction, p i =G·β is the expected allele fraction for locus i, which is a linear function of β, where G is the genotype matrix of n loci and d donors, and β is a vector of length d of unknown contributor fractions. The fifth row of Table 8 shows the data for the third method (NaiveLM).

[0552] It is worth noting that the genotypic information of the contributors was not used to quantify the contributor fraction in the Seq or EPS methods, but it was used in the NaiveLM method. Although the Seq method and the EPS method do not require the use of the genotypic information of the contributors, they produce more reliable results as indicated by the smaller coefficient of variation values ​​compared to the NaiveLM method. In addition, the Seq method and the EPS method have lower bias as indicated by the smaller CVB values ​​compared to the NaiveLM method. The best results among the three methods are marked in bold in Table 8. In short, the two methods using the probability mixture model produce more reliable, accurate and less biased results than the linear regression method.

[0553] Table 8.

[0554]

[0555] Example 2

[0556] Figures 7A-7F The results of the example are shown, and the example uses the data obtained from the actual DNA mixture sample to illustrate that some embodiments can effectively quantify and deconvolve the DNA mixture sample. This example shows that some embodiments can provide improved signal levels for the DNA mixture deconvolution. In this example, narrow prior analysis data is used.

[0557] The samples included DNA from two contributors, with each sample having a 60%-40%, 75%-25%, 90%-10% and 95%-5% fraction of the two contributors. The samples were run in triplicate for each of the subjects NA12878 and NA18507.

[0558] Fig. 7A The main contributor fractions (or "main frequencies" as they are called in the figures) quantified by some embodiments are shown. The horizontal axis shows the actual contributor frequencies of the main contributors. The vertical axis shows the main contributor fractions inferred by the probabilistic mixture model (to the nearest 2.5%). The data show that the predictions provided by the probabilistic mixture model are very close to the true fractions, as shown by the data points located near the identification line.

[0559] Figure 7B The genotypes of the main contributors and minor contributors as predicted by the probability mixture model for the four different alleles in the four subgraphs are shown. The two subgraphs on the left side show the results obtained from the sample of 75-25 contributor scores. The two subgraphs on the right side show the results obtained from the 60-40 score mixture sample. The horizontal axis indicates the markers of different alleles at the locus. The vertical axis shows the allele counts of the locus. All genotypes predicted by the model are correct, with the exception of an allele of the minor contributor at the locus D4S2408 shown in the upper left subgraph. At the locus, the genotype of the real minor contributor is (10, 10), but the model predicts that it is (8, 10). Interestingly, for the minor contributor, the confidence level predicted by the locus is at a relatively low level of 68.6%. In this example, false predictions can be eliminated by setting the detection criteria to be higher than 70%.

[0560] Figure 7CThe number of correct and incorrect calls for contributor genotypes is shown. The horizontal axis shows the actual contributor score of the main contributor (labeled "main frequency" in the figure). The vertical axis shows the number of correct and incorrect calls. The "x" symbol shows the data of the main contributor, and the circle symbol shows the data of the minor contributor. The black symbols show the correct calls, while the gray symbols represent the data of incorrect calls. The horizontal line at 28 indicates the theoretical maximum correct call. Figure 7C Data for calls made at a 90% confidence threshold are shown. Figure 7C The data show that the number of correct detections is relatively high among different contributor scores, while the number of false detections is relatively low and is always below 5. Figure 7C It is also shown that as the contributor fraction increases from 60% to 95%, the correct requirement for top contributors increases and approaches the theoretical maximum.

[0561] Fig.7D Correct and incorrect calls using the same data but with a higher call standard of 99% are shown. Again, across the different major contributor scores, the number of correct calls is consistently high, while the number of incorrect calls is relatively low and always below 5. Fig.7D The detection standard in Figure 7C , so both correct and incorrect detections have lower numbers. However, the correct detections of the main contributors of the 90% and 95% scores are still high and close to the theoretical maximum.

[0562] Fig. 7E Shown with Figure 7C and 7D Similar data, except that the detection threshold was increased to 99.9%. Due to the higher detection threshold, the number of true and false detections is slightly lower than Fig.7D Importantly, there were no false calls at this confidence level, with the exception of three loci that deviated from the model assumptions for known reasons. These loci could have been avoided in the analysis. Other than these cases, the model never attributed high confidence to a false call. Thus, it was appropriately quantifying the uncertainty in the genotype calls.

[0563] Figure 7C-7E The results show that the probabilistic mixture model can accurately determine the genotypes of the contributors. Based on the different requirements in different applications, different detection standard values ​​can be used to achieve the required sensitivity and selectivity.

[0564] Figure 7FShows the correct and incorrect number of calls for whether a DNA mixture sample contains a known contributor's DNA. The horizontal axis shows the actual contributor score (labeled "true major frequency") for the major contributor. The vertical axis shows the evidence value for each locus that the sample includes a genotype. The "x" symbol shows the data for the major contributor, and the circle symbol shows the data for the minor contributor. Figure 7F The data show that there is a relatively high level of evidence that the sample includes two contributors. Not surprisingly, when the scores for the major contributor are 90% and 95%, the level of evidence for the minor contributor is relatively low.

[0565] Example 3

[0566] Figures 8A-8D Results are shown using simulated data to illustrate an example of how some embodiments can effectively quantify and deconvolute a DNA mixture sample. This example shows that some embodiments can provide improved signal levels for deconvoluting a DNA mixture.

[0567] The simulation had four different designs: easy 2-contributors (80-20); difficult - 2 contributors (55-45); easy 3-contributors (60-30-10); and difficult 3-contributors (50-30-20). The easy design had a larger contributor score than the difficult design.

[0568] The simulation included data for 50 loci and 6 alleles. Allelic balance depends on: the number of molecules in the original sample (fixed: 6000), the average molecule detection rate (fixed to 10%; i.e., a locus with 600 molecules detected on average), the allele-to-allele variation in the molecule detection rate (over a range of variations), and sampling noise. The stutter rate was simulated to be 1% or 2%, and the dropout rate to be 1%. A wide range of priors was assumed.

[0569] For the easy 3-contributor (60-30-10) mixture sample, all contributor frequencies were inferred to be within 2.5% of the true value. For the difficult 3-contributor (50-30-20) mixture sample, all contributor frequencies were inferred to be within 7.5% of the true value.

[0570] Figures 8A-8D Data for a simple 3-contributor (60-30-10) mixture sample is shown. Fig. 8A The number of correct and incorrect calls for the contributor genotype is shown. The horizontal axis shows the allele balance of the main contributors. The vertical axis shows the number of correct and incorrect calls. The black symbols show the data for the 1% stutter rate, and the gray symbols show the data for the 2% stutter rate. The solid line shows the correct calls, while the dotted line shows the data for incorrect calls. Fig. 8AData for calls made at a 90% confidence threshold are shown. Fig. 8A The data show that the number of correct calls at different allele balance values ​​is relatively high, while the number of false calls is always close to zero.

[0571] Figure 8B Shows the use of Fig. 8A Same data but with a higher detection standard of 99% correct and incorrect detections. The number of correct detections is significantly lower than Fig. 8A The number of false positives has been minimized, indicating that the 99% threshold may be too stringent in this application. Figure 8C Shown with Fig. 8A and 8B Similar data, except that the detection threshold is increased to 99.9%. The number of correct detections decreases further due to the higher detection threshold. Figure 3 The results of A-3C show that the probabilistic mixture model can accurately determine the genotypes of the contributors, and in this example, a suitable threshold can be set at close to 90% or less than 99%.

[0572] Fig.8D The number of correct and incorrect calls about whether one of the three contributors is included in the DNA mixture sample is shown. The horizontal axis shows the allele balance of the main contributor. The vertical axis shows the evidence value for each locus that the sample contains the genotype. The solid line shows 1% of the Caton error data, and the dotted line shows 2% of the Caton error data. Three different shades of gray show the data of three different contributors. Fig.8D The data show that there is relatively high level evidence for the three contributors across samples including both stutter error conditions.

Claims

1. A method implemented at a computer system comprising one or more processors and a system memory for quantifying a nucleic acid sample comprising nucleic acids of two or more contributors, the method comprising: (a) mapping nucleic acid sequence reads generated from the nucleic acid sample to one or more alleles at one or more polymorphic loci on a reference sequence to identify reads among the nucleic acid sequence reads that match any sequence of a plurality of unbiased target sequences; (b) determining, using the nucleic acid sequence reads and by the one or more processors, an allele count for each of the one or more alleles at the one or more polymorphic loci; (c) applying a probabilistic mixture model to the allele counts, wherein the probabilistic mixture model models the allele counts at the one or more polymorphic loci using a probability distribution, the probability distribution comprising: a probability distribution that accounts for errors in the nucleic acid sequence reads, a probability distribution that accounts for errors in nucleic acid extraction, and a probability distribution that accounts for errors in nucleic acid amplification, and wherein the probabilistic mixture model prunes genotype configurations from data used to quantify nucleic acid fractions of two or more contributors in the nucleic acid sample, wherein the pruned genotype configurations comprise loci that account for all desired alleles by constructing a list of desired alleles and excluding contributors that do not meet a preset number; (d) quantifying, by the one or more processors, one or more fractions of nucleic acid from the two or more contributors in the nucleic acid sample using the probabilistic mixture model; and (e) determining a posterior probability that a particular contributor among the two or more contributors has a particular genotype at one or more polymorphic loci, comprising: (i) multiplying a prior probability of a genotype configuration by the likelihood of the genotype configuration, wherein the prior probability is based on population allele frequency data of the genotype configuration; (ii) normalizing the product of (i) by a sum over genotype space, wherein the genotype space is based on the likelihood genotype as a function of the relationship between two or more contributors; and (iii) summing the genotype configurations that include the specific genotype to obtain the posterior probability.

2. The method according to claim 1, wherein: One or more genotypes of the two or more contributors are unknown.

3. The method according to claim 2 further includes determining one or more allele configurations at each of the one or more polymorphic loci, each allele configuration comprising the allelic states of two or more alleles for each of the two or more contributors.

4. The method of claim 3, further comprising determining an estimated probability of the one or more allele configurations.

5. The method according to claim 1, wherein: The specific genotype comprises a plurality of locus genotypes, and the method further comprises: Across all contributors, summing the posterior probability that a contributor has the particular genotype at all loci; and The occurrence of the designated plurality of loci genotypes in any contributor is determined based on the summed probabilities.

6. The method according to claim 5, wherein: The nucleic acid sample is a forensic sample and the data of genotypes of the plurality of loci is obtained from a person of interest, the method further comprising determining that the person of interest is a contributor to the nucleic acid sample.

7. The method according to claim 1, wherein: The nucleic acid sample comprises DNA molecules and / or RNA molecules.

8. The method according to claim 7, wherein: The nucleic acid sequence reads are obtained by sequencing the DNA molecules and / or RNA molecules using a unique molecular index.

9. The method according to claim 1, wherein: The probability distribution comprises a first binomial distribution.

10. The method according to claim 9, wherein: The first binomial distribution is expressed as follows: n ij ~BN(n i ,p ij ) in n ij is the allele count of allele j at locus i; n i is the total allele count at locus i; as well as p ij is a probability parameter indicating the probability of allele j at locus i.

11. The method according to claim 10, wherein: The probability parameter p ij is the following function: (i) the fraction of nucleic acid in the nucleic acid sample that is one of the two or more contributors, or β; (ii) the genotypes of the two or more contributors, or G; and / or (iii) errors in the reads of the nucleic acid sequence, or θ.

12. The method according to claim 11, wherein: The probabilistic mixture model uses a Beta distribution to model the errors in the nucleic acid sequence reads.

13. The method according to claim 12, wherein: The Beta distribution is defined by a mean parameter μ and a concentration parameter k.

14. The method according to claim 13, wherein: The concentration parameter k has priors representing different noise conditions, and the concentration parameter varies across the loci.

15. The method according to claim 12, wherein: The probability mixture model uses the combined first binomial distribution and the beta distribution to obtain n following a beta-binomial distribution. ij to produce a combined error model that models errors in nucleic acid sequence reading and errors in DNA extraction of nucleic acid samples.

16. The method according to claim 15, wherein: The Beta-Binomial distribution has the following form: BB(n ij |n i , m, k).

17. The method according to claim 1, wherein: (d) comprising quantifying one or more fractions of nucleic acids of the two or more contributors in the nucleic acid sample by maximizing a likelihood function of the nucleic acid sequence reads.

18. The method according to claim 17, wherein: (d) include: calculating a plurality of likelihood values ​​using the plurality of potential score values ​​and the likelihood function of the allele count determined in (b); identifying a latent score vector associated with a maximum likelihood value; and The one or more fractions of nucleic acid of the two or more contributors in the nucleic acid sample are quantified using the identified potential fraction vectors.

19. The method according to claim 17, wherein: The likelihood function depends on P(G|π), which is the prior probability of the genotypes of the two or more contributors given the population allele frequencies (π).

20. The method according to claim 19, wherein: The prior probability P(G|π) is calculated using a marginal distribution that satisfies the Hardy-Weinberg equilibrium.

21. The method according to claim 19, wherein: The prior probabilities are calculated considering virtual alleles with fixed prior probabilities representing mechanism dropout.

22. The method according to claim 9, wherein: The probabilistic mixture model uses the second binomial distribution to model stuttering errors in allele data.

23. The method according to claim 22, wherein: The second binomial distribution is expressed as follows: s ik ~BN(n i(k+1) ,r i ) in s ik is the Caton allele count at locus i of the Caton allele that appears to be allele k but is actually caused by a Caton error of allele k+1; n i(k+1) is the raw allele count of allele k+1 at locus i; and r i is the stuttering rate of locus i.

24. The method according to claim 23, wherein: The stutter rate r varies across the loci and has priors representing different noise conditions that are shared across the loci.

25. A method according to claim 23, wherein the probabilistic mixture model uses a likelihood function to quantify the nucleic acid fractions of the two or more contributors in the nucleic acid sample, and the likelihood function includes the product of the likelihood of non-catton allele counts and the likelihood of catton allele counts.

26. The method of claim 23, wherein: When determining the number of molecules that potentially cause jamming, the probabilistic mixture model adds a fixed number of molecules to the allele count assigned to allele k+1.

27. The method of claim 1, wherein: The probabilistic mixture model uses virtual out-of-sample alleles to model natural dropout.

28. The method according to claim 27, wherein: The prior for the virtual out-of-sample alleles is proportional to the number of unobserved alleles.

29. The method according to claim 28, wherein: The number of unobserved alleles is estimated by: interpolate all integers between the shortest and longest observed integer-valued alleles, Add any observed non-integer-valued alleles, and Returns the maximum result value and the threshold value.

30. The method of claim 1, wherein: The list of desired alleles consists of alleles that have allele counts above the threshold and allele counts that are too high to be plausible due to stuck insertions.

31. The method according to claim 30, wherein: The threshold is the sum of: (i) the maximum non-catton allele count, and (ii) the value multiplied by the potential catton donor allele count.

32. The method of claim 1, wherein: Pruning genotype configurations involves removing genotype configurations that have a mismatch between the allele data and the expected allele counts.

33. The method of claim 32, wherein: The genotype configurations having mismatches have root mean square error (RMSE) values ​​greater than one or more thresholds.

34. The method of claim 1, wherein: The alleles at the one or more polymorphic loci comprise single nucleotide polymorphism (SNP) alleles and / or short tandem repeat (STR) alleles.

35. A computer system comprising a system memory and one or more processors, configured to: (a) mapping nucleic acid sequence reads generated from a nucleic acid sample to one or more alleles at one or more polymorphic loci on a reference sequence to identify reads among the nucleic acid sequence reads that match any sequence of a plurality of unbiased target sequences; (b) determining an allele count for each of the one or more alleles at the one or more polymorphic loci using the nucleic acid sequence reads; (c) applying a probabilistic mixture model to the allele counts, wherein the probabilistic mixture model models the allele counts at the one or more polymorphic loci using probability distributions, the probability distributions comprising: a probability distribution that accounts for errors in the nucleic acid sequence reads, a probability distribution that accounts for nucleic acid extraction errors, and a probability distribution that accounts for nucleic acid amplification errors, and wherein the probabilistic mixture model prunes genotype configurations from data used to quantify nucleic acid fractions of two or more contributors in the nucleic acid sample, wherein the pruned genotype configurations comprise loci that account for all desired alleles by constructing a list of desired alleles and excluding contributors that do not meet a preset number; (d) quantifying one or more fractions of nucleic acid from the two or more contributors in the nucleic acid sample using the probabilistic mixture model; as well as (e) determining a posterior probability that a particular contributor among the two or more contributors has a particular genotype at one or more polymorphic loci, comprising: (i) multiplying a prior probability of a genotype configuration by the likelihood of the genotype configuration, wherein the prior probability is based on population allele frequency data of the genotype configuration; (ii) normalizing the product of (i) by a sum over genotype space, wherein the genotype space is based on the likelihood genotype as a function of the relationship between two or more contributors; and (iii) summing the genotype configurations that include the specific genotype to obtain the posterior probability.

36. The system of claim 35, further comprising means for extracting nucleic acid from the nucleic acid sample.

37. The system of claim 35, wherein the one or more processors are further configured to determine a total number of contributors among the two or more contributors.

38. A system according to claim 35, wherein the one or more processors are further configured to determine an allelic configuration at each of the one or more polymorphic loci, the allelic configuration comprising the allelic states of two or more alleles for each of the two or more contributors.

39. A non-transitory computer readable medium storing program code, which, when executed by one or more processors of a computer system, causes the computer system to implement a method for quantifying a nucleic acid sample comprising nucleic acids of two or more contributors, the program code comprising: (a) code for mapping nucleic acid sequence reads generated from a nucleic acid sample to one or more alleles at one or more polymorphic loci on a reference sequence to identify reads among the nucleic acid sequence reads that match any sequence of a plurality of unbiased target sequences; (b) code for determining an allele count for each of the one or more alleles at the one or more polymorphic loci using the nucleic acid sequence reads; (c) code for applying a probabilistic mixture model to the allele counts, wherein the probabilistic mixture model uses probability distributions to model the allele counts at the one or more polymorphic loci, the probability distributions comprising a probability distribution that accounts for errors in the nucleic acid sequence reads, a probability distribution that accounts for nucleic acid extraction errors, and a probability distribution that accounts for nucleic acid amplification errors, and wherein the probabilistic mixture model prunes genotype configurations from data used to quantify nucleic acid fractions of two or more contributors in the nucleic acid sample, wherein the pruned genotype configurations comprise loci that account for all desired alleles by constructing a list of desired alleles and excluding contributors that do not meet a preset number; (d) code for quantifying one or more fractions of nucleic acids of the two or more contributors in the nucleic acid sample using the probabilistic mixture model; as well as (e) code for determining a probability that a particular contributor among the two or more contributors has a particular genotype at one or more polymorphic loci, wherein determining the posterior probability that a particular contributor among the two or more contributors has a particular genotype at one or more polymorphic loci comprises: (i) multiplying the prior probability of the genotype configuration by the likelihood of the genotype configuration, where The prior probability is based on population allele frequency data of genotype configuration; (ii) normalizing the product of (i) by a sum over a genotype space, wherein the genotype space is based on the likelihood genotype as a function of the relationship between two or more contributors; as well as (iii) summing the genotype configurations that include the specific genotype to obtain the posterior probability.

Citation Information

Patent Citations

  • Methods and apparatus for measuring analytes using large scale FET arrays

    US20090026082A1

  • Methods for sequencing polynucleotides

    US20120053063A1

  • Detecting and classifying copy number variation

    US20130029852A1

  • Error suppression in sequenced DNA fragments using redundant reads with unique molecular indices (UMIS)

    US20160319345A1

  • Paired end sequencing

    US7601499B2