Systems and methods for variant detection

By improving the variant detection algorithm and combining linear models and machine learning classifiers, the accuracy and specificity issues of variant detection in existing technologies have been resolved. This enables efficient identification of single nucleotide variants and insertion/deletion variants in nanopore sequencing data, reducing false positive and false negative rates.

CN123125653APending Publication Date: 2026-11-17ROCHE SEQUENCING SOLUTIONS INC
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202580028621.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-04-17
Filing Date
2025-04-17
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing sequencing technologies struggle to efficiently and accurately identify single nucleotide variants and insertion/deletion variants in DNA sequencing data, especially in newer next-generation sequencing technologies such as nanopore sequencing data, where existing tools fail to function well, resulting in high false positive and false negative rates.

Method used

A variant detection algorithm is employed, comprising barcode deduplication, preprocessing, hotspot modules, and an adaptive module. It utilizes a core linear model and a machine learning classifier to identify variants through a two-stage approach. The algorithm performs secondary analysis on nanopore sequencing data, employing a linear model and a nonlinear tree-based ensemble machine learning classifier, combined with multiple features, for variant detection.

Benefits of technology

It improves the accuracy and specificity of variant detection, reduces false positives and false negatives, and can better identify low-support variants, making it suitable for variant detection in tissue and plasma samples.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123125653A_ABST
    Figure CN123125653A_ABST
Patent Text Reader

Abstract

The variant detector algorithm described herein can be used to identify various mutations in sequencing data. Depending on the implementation, the InDel detector and / or SNV detector can use a multi-level filtering approach involving a core linear model and additional models to provide additional filtering of identified variants, thereby improving sensitivity and / or accuracy. A barcode deduplication module can also be used to demultiplex multiplexed samples, allowing the variant detector algorithm to be applied individually to each sample. Additional filtering steps can be used to further reduce false positives and / or false negatives.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Cross-references to related applications

[0002] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 635,576, filed April 17, 2024, which is hereby incorporated in its entirety by reference. Background Technology

[0003] The development of affordable and rapid DNA sequencing technologies has enabled the development of targeted therapies, which rely on the use of DNA biomarkers to identify patients suitable for targeted treatment. For example, mutations in certain genes, such as those involved in cell proliferation, are known to cause certain types of cancer that can be treated very effectively with specific types of drugs. Other mutations are known to confer resistance to certain therapies. Therefore, there is a need to improve the systems and methods used to identify variants from sequencing data. Summary of the Invention

[0004] The embodiments described herein relate to systems and methods for detecting variants in sequencing data. More specifically, the embodiments described herein relate to detecting single nucleotide variants, insertions, and deletions in sequencing data.

[0005] According to a first aspect of this disclosure, a method is provided for detecting insertions and deletions (InDels) from sequencing data generated from multiple samples. The method includes receiving a computer file containing multiple paired sequence reads; aligning the multiple paired sequence reads with a reference sequence; sorting the aligned and paired sequence reads; grouping the sorted, aligned, and paired sequence reads into multiple groups based on a unique molecular identifier (UMI), a read start position, and a read end position, wherein each member of a group has the same UMI, read start position, and read end position; determining a consensus sequence from the members of each group; parsing the paired consensus sequence reads; correcting the parsed consensus sequence reads; determining a set of candidate InDel variants based on a comparison of the corrected consensus sequence reads with the reference sequence; for each candidate InDel variant, calculating a weighted linear score using a linear model, wherein a candidate InDel variant is identified as a potential InDel variant when the weighted linear score is greater than a linear model threshold; and for each potential InDel variant, calculating a score using a nonlinear tree-based ensemble machine learning classifier, wherein a potential InDel variant is identified as a true InDel variant when the score is greater than a machine learning classifier threshold.

[0006] In some embodiments of the first aspect, the method further includes generating a computer file containing paired consensus sequence read segments.

[0007] In some embodiments of the first aspect, the step of correcting the parsed consensus sequence reads includes: comparing the parsed consensus sequence reads with a reference sequence to determine a set of preliminary InDel variants, including a count of each preliminary InDel variant; and adjusting the count of each preliminary InDel variant based on a comparison of the count of each preliminary InDel variant with a corresponding background variant count determined from a control sample.

[0008] In some embodiments of the first aspect, the comparison includes determining whether the count of each preliminary InDel variant is background noise.

[0009] In some embodiments of the first aspect, the step of determining whether the count of each initial InDel variant is background noise includes calculating the probability that the initial InDel variant is background noise.

[0010] In some embodiments of the first aspect, the linear model includes double-chain mutation count, single-chain mutation count, and consensus mutation count.

[0011] In some embodiments of the first aspect, the threshold of the linear model is sample-specific.

[0012] In some embodiments of the first aspect, for all candidate InDel variants in a sample, a threshold for the linear model is determined by performing a regression fit on a graph of the cumulative candidate InDel variant counts as a function of a weighted linear score, and setting the threshold at a weighted linear score that produces a target cumulative InDel variant count having at least one weighted linear score.

[0013] In some embodiments of the first aspect, a threshold for a linear model is determined from a set of training data by determining the cumulative count of false positive variant detections as a function of a weighted linear score, and selecting a weighted linear score corresponding to a target cumulative false positive count as the threshold.

[0014] In some embodiments of the first aspect, the nonlinear tree-based ensemble machine learning classifier includes single-stranded molecule counts, consensus molecule counts, double-stranded molecule counts, the ratio of single-stranded to double-stranded molecule counts, the median distance from the 5' end of a potential variant distance sequence read, the relative diffusion of variant distances in variant molecules relative to the background, the relative diffusion of insertion start positions in variant molecules relative to the background, the original alignment chain bias of consensus molecules, the original origin chain bias of single-stranded molecules, the original alignment chain bias of non-double-stranded molecules, the proportion of double-stranded variant molecules with two reads, the proportion of double-stranded variant molecules with three or fewer reads, the proportion of consensus variant molecules with two reads, the proportion of consensus variant molecules with three or fewer reads, the relative frequency of single-stranded molecules in variant molecules relative to the background, the relative frequency of double-stranded molecules in variant molecules relative to the background, and the 2 x 2 Fisher exact test p-value of the ratio of single-stranded to double-stranded molecules in variant molecules relative to the background.

[0015] In some embodiments of the first aspect, a set of healthy samples is used to calibrate the threshold of a machine learning classifier to achieve a predetermined false positive rate.

[0016] According to a second aspect of this disclosure, a method is provided for detecting single nucleotide variants (SNVs) from sequencing data generated from multiple samples. The method includes: receiving a computer file containing multiple paired sequence reads; aligning the sequence reads with a reference sequence; sorting the aligned sequence reads by pairing; grouping the sorted, aligned, and paired sequence reads into multiple groups by unique molecular identifiers (UMIs), read start positions, and read end positions, wherein each member of a group has the same UMI, read start position, and read end position; determining a consensus sequence from the members of each group; parsing the paired consensus sequence reads; correcting the parsed consensus sequence reads; determining a set of candidate variants based on comparisons of the corrected consensus reads; for each candidate SNV variant, calculating a weighted linear score using a linear model, wherein a candidate SNV variant is identified as a potential SNV variant when the weighted linear score is greater than a linear model threshold; and for each potential SNV variant, calculating a weighted logistic score using a logistic regression model, wherein a potential SNV variant is identified as a true variant when the weighted logistic score is greater than a logistic regression model threshold.

[0017] In some embodiments of the second aspect, the method further includes generating a computer file containing paired consensus sequence read segments.

[0018] In some embodiments of the second aspect, the step of correcting the parsed consensus sequence reads includes: comparing the parsed consensus reads with a reference sequence to determine a set of preliminary SNV variants, including a count of each preliminary SNV variant; and adjusting the count of each preliminary SNV variant based on a comparison of the count of each preliminary SNV variant with a corresponding background variant count determined from a control sample.

[0019] In some embodiments of the second aspect, the comparison includes determining whether the count of each initial SNV variation is background noise.

[0020] In some embodiments of the second aspect, the step of determining whether the count of each initial SNV variation is background noise includes calculating the probability that the initial SNV variation is background noise.

[0021] In some embodiments of the second aspect, the linear model includes double-chain mutation count, single-chain mutation count, and consensus mutation count.

[0022] In some embodiments of the second aspect, the threshold of the linear model is sample-specific.

[0023] In some embodiments of the second aspect, a threshold for the linear model is determined for all candidate SNV variants in the sample by performing a regression fit on a graph of the cumulative candidate SNV variant counts with at least some weighted linear fraction as a function of the weighted linear fraction and setting a threshold to the weighted linear fraction that produces the target cumulative SNV variant count.

[0024] In some embodiments of the second aspect, the threshold of the linear model is determined from a set of training data by determining the cumulative count of false positive variant detections as a function of a weighted linear score, and selecting the weighted linear score corresponding to a target cumulative false positive count as the threshold.

[0025] In some embodiments of the second aspect, the logistic regression model includes double-chain mutation count, single-chain mutation count, and consensus mutation count.

[0026] In some embodiments of the second aspect, the logistic regression model further includes the ratio of single-stranded variant counts to double-stranded molecule counts, adjustments for coverage depth, adjustments for errors in the gene, median distance of the variant from the 5' end of the sequence read, relative diffusion of variant distance in the variant molecule relative to the background, relative diffusion of insertion start position in the variant molecule relative to the background, relative diffusion of insertion end position in the variant molecule relative to the background, original alignment strand bias, original origin strand bias, type of substitution of the variant, two base pairs of context surrounding the variant position, proportion of double-stranded variant molecules with two reads, proportion of double-stranded variant molecules with three or fewer reads, proportion of consensus variant molecules with two reads, proportion of consensus variant molecules with three or fewer reads, relative frequency of consensus in the variant molecule relative to the background, and relative frequency of double-stranded variants in the variant molecule relative to the background.

[0027] In some embodiments of the second aspect, the logistic regression model includes multiple features, each of which is multiplied by a weight, wherein the weights are determined from a set of training data having known true positive variants and known false positive variants.

[0028] According to a third aspect of this disclosure, a system for generating sequencing data is provided. The system may include a measurement device and / or a logic system. The logic system may include a processor coupled to a memory storing instructions executable by the processor. When executing the instructions, the processor is configured to perform the methods of one or more embodiments of the first aspect.

[0029] In some embodiments of the third aspect, the system further includes a treatment device for determining treatment or administering treatment to a patient based on at least one true InDel variant.

[0030] In some embodiments of the third aspect, the system further includes a reporting device for displaying information about at least one genuine InDel variant.

[0031] According to a fourth aspect of this disclosure, a non-transitory computer-readable medium is provided. The computer-readable medium stores a set of instructions that, when executed by at least one processor, cause the processor to perform one or more embodiments of the first aspect.

[0032] According to a fifth aspect of this disclosure, a system for generating sequencing data is provided. The system may include a measurement device and / or a logic system. The logic system may include a processor coupled to a memory storing instructions executable by the processor. When executing the instructions, the processor is configured to perform the methods of one or more embodiments of the second aspect.

[0033] In some embodiments of the fifth aspect, the system further includes a treatment device for determining treatment or administering treatment to a patient based on at least one real variant.

[0034] In some embodiments of the fifth aspect, the system further includes a reporting device for displaying information about at least one real variant.

[0035] According to a sixth aspect of this disclosure, a non-transitory computer-readable medium is provided. The computer-readable medium stores a set of instructions that, when executed by at least one processor, cause the processor to perform one or more embodiments of the second aspect. Attached Figure Description

[0036] The features of the invention are specifically set forth in the appended claims. The features and advantages of the invention will be better understood by referring to the following detailed description of illustrative embodiments (in which the principles of the invention are utilized) and the accompanying drawings.

[0037] Figure 1 is a flowchart of a method for the workflow of an InDel detector algorithm according to some embodiments.

[0038] Figure 2 shows an overview of barcode deduplication algorithms according to some embodiments.

[0039] Figure 3 is a flowchart of a method for a barcode deduplication algorithm according to some embodiments.

[0040] Figure 4 is another flowchart of a method for a barcode deduplication algorithm according to some embodiments, which illustrates the creation of a barcode family.

[0041] Figure 5 illustrates how the threshold of the core linear model is set according to some embodiments.

[0042] Figure 6 is a flowchart of a method for a single nucleotide variant (SNV) detector algorithm according to some embodiments.

[0043] Figure 7 illustrates the improved performance of a two-stage approach for an SNV detector algorithm according to some embodiments.

[0044] Figure 8 illustrates a sequencing system according to some embodiments.

[0045] Figure 9 illustrates an exemplary computer device according to some embodiments. Detailed Implementation

[0046] This disclosure provides numerous techniques for variant detection as part of a secondary analysis workflow of sequencing data generated by today's next-generation sequencing devices. More specifically, a variant detection algorithm is provided for identifying single nucleotide variants (SNVs) and / or insertions / deletions (InDels) from sequencing data, such as nanopore-based sequencing data.

[0047] In at least one aspect of a secondary analysis workflow, an InDel detector algorithm is provided. The InDel detector algorithm may include a hotspot module and an adaptive module. In the hotspot module, the InDel detector algorithm uses numerous heuristics to identify InDel variants in hotspot regions. Simultaneously, the adaptive module uses a two-stage approach to identify variants with low AF (Aspect-Oriented Fissure) in plasma samples that may not be strongly supported. In the first stage, multiple variants are filtered out using a core linear regression model based on a comparison of the score of each candidate variant with a threshold. In the second stage, each variant that passes the first stage is scored using a machine learning classifier (such as a Gradient Boosting Machine (GBM) model), and the classifier score is then compared with a threshold to determine a final list of detected variants. Optionally, the detected variants can be further filtered by a blocking list filter.

[0048] In another aspect of the secondary analysis workflow, an SNV detector algorithm is provided, which is similar in many ways to the InDel detector algorithm described above. The SNV detector algorithm may include a hotspot module and an adaptive module. The hotspot module is similar to the hotspot module of the InDel detector algorithm, except that it may include one or more additional heuristics in addition to or instead of the heuristics used in the InDel detector algorithm. The adaptive module of the SNV detector algorithm also uses a two-stage approach to identify variants. In the first stage, multiple variants are filtered out using a core linear regression model based on a comparison of the score of each candidate variant with a threshold. In the second stage, variants that passed the first stage are filtered out using an extended linear model that incorporates multiple additional weighted features, based on a comparison of the score of each candidate variant with a threshold.

[0049] These secondary analysis workflows can be used to analyze sequencing data from many next-generation sequencing devices, including nanopore-based sequencing systems. Correctly identifying variants from raw sequencing data is a crucial step in bioinformatics processes used for patient diagnosis and / or the creation of disease-specific tailored therapies.

[0050] sequencing

[0051] Sequencing analysis can be used as part of a procedure to determine sequencing reads from multiple microsatellite loci, allowing the sequencing of prepared target nucleic acid molecules (e.g., sequencing libraries). A variety of sequencing technologies or sequencing analyses can be used. As used herein, the term "next-generation sequencing (NGS)" refers to a sequencing method that allows for the massive parallel sequencing of cloned amplified molecules and individual nucleic acid molecules (or nucleic acid analogs).

[0052] Non-limiting examples of sequencing assays suitable for use with the methods disclosed herein include nanopore sequencing (see, for example, U.S. Patent Application Publication Nos. 2013 / 0244340, 2013 / 0264207, 2014 / 0134616, 2015 / 0119259, and 2015 / 0337366), Sanger sequencing, capillary array sequencing, thermal cycling sequencing (see, for example, Sears et al., Biotechniques, 13:626-633 (1992)), solid-phase sequencing (see, for example, Zimmerman et al., Methods Mol. Cell Biol., 3:39-42 (1992)), and sequencing using mass spectrometry such as matrix-assisted laser desorption / ionization time-of-flight mass spectrometry (MALDI-TOF / MS) (see, for example, Fu et al., Nature Biotech., 16:381-384). (1998)), sequencing-by-hybridization (see, for example, Drmanac et al., Nature Biotech., 16:54-58(1998)) and NGS methods, including but not limited to sequencing-by-synthesis (see, for example, HiSeq™, MiSeq™ or Genome analyzers, each available from Illumina, Inc. (San Diego, CA)), sequencing-by-ligation (see, for example, SOLiD™, available from Thermo Fisher Scientific, Inc. (Waltham, MA)), ion semiconductor sequencing (see, for example, Ion Torrent™, available from Thermo Fisher Scientific, Inc. (Waltham, MA)), and SMRT® sequencing, available from Pacific Biosciences of California, Inc. (MenloPark, CA).

[0053] Commercially available sequencing technologies include hybridization-while-sequencing platforms from Affymetrix, Inc. (Sunnyvale, CA), synthesis-while-sequencing platforms from Illumina, Inc. (San Diego, CA), and ligation-while-sequencing platforms from Thermo Fisher Scientific, Inc. (Waltham, MA). Other sequencing technologies include, but are not limited to, Ion Torrent technology from Thermo Fisher Scientific, Inc. (Waltham, MA) and nanopore sequencing from Roche Sequencing Solutions, Inc. (Santa Clara, CA) and / or Oxford Nanopore Technologies, plc (Oxford, United Kingdom).

[0054] Bioinformatics Workflow Overview

[0055] The output of NGS sequencers is typically processed by a bioinformatics workflow. This workflow processes the raw signal from the NGS sequencer and converts it into base detections, commonly referred to as raw reads. These raw reads are usually stored in a FASTQ file, which combines the raw reads with relevant quality data. This part of the bioinformatics workflow is often referred to as primary analysis.

[0056] The next part of the bioinformatics workflow is called secondary analysis, which takes the raw reads generated by the primary analysis and performs several tasks, including alignment and variant detection.

[0057] Tertiary analysis is the final part of the bioinformatics process and uses variant detection information to generate medical insights that healthcare practitioners can use to improve patient care.

[0058] Secondary analysis

[0059] Sequencing data generated by new sequencing technologies, such as nanopore-based sequencers, exhibits different characteristics compared to sequencing data generated by currently market-leading sequencers, such as Illumina sequencers. These differences may include variations in raw read accuracy and error profile. Because Illumina sequencers currently dominate the market, the vast majority of secondary analysis software tools developed are tailored to handle the data types generated by Illumina sequencers. These tools generally work well with data from Illumina sequencers but may not perform well with data generated by newer next-generation sequencing technologies, such as nanopore sequencers. Therefore, there is a need to develop new secondary analysis tools to work effectively with emerging sequencing technologies currently under development. Furthermore, while the variant detection method described herein may be particularly effective for nanopore sequencing data, it can also be used with other types of sequencing data, such as data from Illumina sequencers.

[0060] InDel detector

[0061] Figure 1 illustrates the secondary analysis workflow for detecting insertions and deletions (InDels). InDel mutations refer to one of the most common short variant categories involving nucleotide insertions and / or deletions in genomic DNA. Somatic InDels refer to insertion and deletion events in non-germline cells and can lead to cancer. These variants constitute an important component of the genetic variation in cancer genomes. Indels in coding regions can disrupt protein coding and often lead to changes in protein function. Given that InDels are less studied than single nucleotide variants (SNVs), refining and optimizing InDel detection tools is important. This article describes improved systems and methods for detecting Indels with improved accuracy and / or specificity, producing fewer false positives and / or fewer false negatives.

[0062] The InDel algorithm workflow can be divided into five main steps or modules: barcode deduplication step or module 100, preprocessing step or module 102, multi-stage InDel detection step or module 104, filtering step or module 106, and output step or module 108.

[0063] The barcode deduplication step is further illustrated in Figures 2 through 4. Figure 2 provides an overview of the barcode deduplication process, which is designed to fold correctly paired reads (including double-stranded reads) with the same fragment start, fragment end, and unique molecular identifier (UMI) into the same barcode family. As shown in Figure 2, paired reads from the positive strand 200 and the negative strand 202 are folded into a single barcode family 204 by the barcode deduplication step because they share the same UMI, fragment start, and fragment end. Consensus detection can be performed within the barcode family to identify candidate variants (SNVs and InDels). The output of the barcode deduplication step is a file (i.e., a BAM file) containing consensus read pairs for each barcode family.

[0064] Figure 3 illustrates the workflow of the barcode deduplication process, beginning with the sequencer output typically provided in a FASTQ file 300. During this process, paired FASTQ files are used for sorting. For example, the FASTQ file contains raw sequencing data, which can be aligned using SWA or other alignment techniques 302. The output of alignment step 302 is typically a BAM file 304, which can be sorted by pair to produce paired BAM files 306. For each read pair, low-quality reads and off-target reads can be filtered out here or in an early stage before barcode families are grouped together 308. After grouping barcode families together 308, consensus detection 310 can be performed on each barcode family group, which produces a deduplicated BAM file 314. The deduplicated BAM file 314 may also include a variant file, a base count file, and / or a metric file 312.

[0065] Figure 4 illustrates an embodiment of the barcode family grouping step. For each record 400, 401, a barcode family key 402 is constructed or extracted from records 400, 401. The barcode family key may be a UMI associated with a paired read segment. The barcode family key 402 is then checked to see if it already exists in the current barcode family link list 404. If the barcode family key 402 already exists, the barcode family information can be added to records 406, 407. If the barcode family key does not exist in the current barcode family link list 404, the segment start position is compared with the previous segment start position 408. If the segment start position is the same as the previous start position, a new barcode family 410 is created, and the new barcode family information is added to records 412, 413 to start the new barcode family. If the fragment starts at a different position than the previous one, the current link list 414 of the barcode family is processed to output the barcode family mutation 416 and the barcode family is output to the deduplication output file.

[0066] In some embodiments, the barcode deduplication step 100 may also optionally exclude any soft clipped segments, as soft clipped segments may be associated with a higher error rate.

[0067] As a result of the barcode deduplication step, multiple samples containing more than one sample that are all mixed together can be demultiplexed into individual samples, allowing each sample to be analyzed as described herein.

[0068] Returning to Figure 1, after the barcode deduplication step 100, preprocessing 102 can be performed. Preprocessing of the deduplicated output can include background correction, where the consensus sequence from the deduplicated output can be compared with a reference sequence from a healthy sample. This can be an independent sample unrelated to the patient sample, or it can be a matched sample where healthy tissue is collected from the patient along with tumor or unhealthy samples from the patient. Comparison with the healthy sample allows filtering out detected variations in both the healthy and actual samples as possible artifacts or errors introduced at some point during the sequencing process, such as, for example, recurring library preparation errors. In some embodiments, background correction is performed only on plasma samples and not on tissue samples. In other embodiments, background correction can be performed on both plasma and tissue samples.

[0069] In some embodiments, the background correction step can be performed by capturing errors (low allele frequency (AF) variants) as systematic errors from a healthy control cfDNA sample set, which are not expected to have low AF variants. A probabilistic model is built from the measured errors at each location in the control reference set to generate a background error distribution. Then, for the test sample, the number of supporting reads at all non-reference locations is evaluated based on the derived background error distribution for each of the non-reference locations. If a variant appears to belong to the background noise distribution, it is "corrected" (i.e., filtered out) from the basic count file. That is, the number of supporting reads for that variant is set to 0. If a variant in the test sample appears to be significantly different from the background error distribution, it is not corrected.

[0070] Furthermore, preprocessing may include parsing the deduplicated output, which may include computing features used in subsequent steps of the InDel detector algorithm. These features include, but are not limited to, those listed in Table 1 below.

[0071] Table 1

[0072]

[0073] These features can be passed to the InDel detection algorithm 104. As shown in Figure 1, the InDel detection algorithm 104 may include multiple components, such as a hotspot module and an adaptive module.

[0074] The hotspot module can be used for predetermined target loci, which can include cancer "hotspots" from the target group, including known operable mutations (e.g., BRAF, V600E, etc.), known biomarkers (e.g., toxicity biomarkers DPYD or resistance biomarkers), and recurrent cancer mutations (e.g., the top 1% of most mutated mutations in COSMIC). The variant detector can be designed to be more sensitive to mutations on the hotspot list.

[0075] Hotspot modules can use heuristic rules to identify variants. For example, for tissue samples, variants are detected when either the double-strand rule or the depth rule is met. The double-strand rule is met when the AF is greater than or equal to 1.5%, the double-strand support is greater than or equal to 5, and the total support is greater than or equal to the double-strand support, where support is the number of reads with that variant. The depth rule is met when the AF is greater than 1.5% and the variable depth is greater than 5.

[0076] For cfDNA (i.e., plasma samples), the duplex rule is satisfied when (1) the duplex support is greater than or equal to the minimum duplex support, and (2) the total support is greater than or equal to the duplex support. The depth rule is satisfied when the variable depth is greater than 5.

[0077] An adaptive module can be used in parallel with the heuristic hotspot module. The adaptive module can include a two-stage approach. The first stage uses a core linear model to compute a weighted variable molecule count and learns a threshold for sample specificity for detection. The second stage applies an additional filter based on a LightGBM (gradient-enhanced decision tree) classifier, which uses both the variable molecule count and additional techniques and biological features (as listed in Table 1). Compared to other standard methods, this enables the detector to detect low allele proportions of variants with limited support in cfDNA samples.

[0078] The core linear model of the first stage is described in detail below. For a specific candidate variant identified by the deduplication step, , , They represent double strands ( ),consensus( ) and single chain ( ) Variant reads. The goal of this model is to determine the optimal coefficient α ( ), β ( ) and γ ( To minimize the total error (false positives and false negatives), a weighted linear model is used as follows:

[0079]

[0080] in For double-stranded variant counting, For consensus mutation count, and This is for counting mutations in a single-stranded organism. Theoretically, when the weighted linear fraction... If the value exceeds a certain cutoff, the mutation will be detected. Therefore, each mutation will receive a weighted linear score according to this linear model.

[0081] The cutoff value for each sample is learned in a sample-specific manner by fitting regressions to all candidate variants in the sample. This consists of the following steps:

[0082] 1. Cumulative count(X) = Total variation with weighted linear fractions (F(d, c, s)) >= X

[0083] 2. Fitting the regression log(cumulative count(X)) ~ log(X)

[0084] 3. Set a threshold for when the fit_cumulative_count / group size drops to a low value, for example, 1 / 10000 bp.

[0085] Figure 5 shows the threshold as a vertical line on the graph.

[0086] The second-stage machine learning classifier is described in detail below. All candidate variants identified by the core linear model in the first stage are scored based on a machine learning classifier, such as a non-linear tree-based ensemble learning classifier (e.g., LightGBM). Table 1 above describes the complete list of features used by the LightGBM classifier. Variants from the first stage whose classifier scores are above a certain threshold are retained as final detections. The classifier's threshold is calibrated using a set of healthy samples to achieve the desired or predetermined false positive level. Other types of classifiers can be used, preferably those with interpretable features.

[0087] Returning to Figure 1, the output 106 from the InDel detector can then be filtered using a block list filter and other filter labels. A block list filter can be generated to block repetitive false positive variants detected in multiple cohorts. Variants included in the block list filter can be limited to false positive variants at the somatic AF (AF < 30%) level, where the false positive variant must be found in more than two samples in the same cohort and in two or more independent cohorts. Other filter labels that can be applied include a low AF label applied when the variant AF is less than a minimum AF threshold, a low support label applied when the variant support is less than a minimum support threshold, and a low depth label applied when the variant depth is less than a minimum depth threshold. Applying these labels or block list filters to variants will remove variants from the list of detected variants.

[0088] After filtering 106, for example, the output is a list of detected variants that can be written into a VCF file.

[0089] SNV detector

[0090] A single nucleotide variant (SNV) is a mutation with a single change in the A / G / C / T bases. Somatic SNVs occur only in somatic cells (non-germline cells). Cancer mutations fall into this category. In contrast, another type of SNV is germline or hereditary because it occurs in sperm / egg cells. This type of mutation is called an SNP (single nucleotide polymorphism). The SNV detector algorithm described herein was developed for detecting SNVs in cfDNA and tissue samples, and its primary goal is to detect any variants observed in the data and distinguish them from technical artifacts. Variants can be further classified by type (i.e., somatic or germline) in downstream modules. The SNV detector module described here and shown in Figure 6 comprises three main components, as described below.

[0091] (1) Preprocessing module 602 performs calculations and operations on the barcode deduplication output file 600 before the main body of SNV detector 604, including parsing the variant file and depth file, where depth refers to the number of times a genomic locus has been sequenced. The SNV detector also incorporates background correction. As described above, for the InDel detector, preprocessing includes calculating the features used in the SNV detection algorithm.

[0092] (2) The SNV detector 604 consists of several components: a core linear model, an adaptive detector, a hotspot detector, and a final filter based on an extended linear model.

[0093] The adaptive detector and the hotspot detector are two parallel detection modules. The adaptive detector detects variants using a coverage selector approach. In the adaptive detector, the error distribution for each of the 12 substitution types or 192 trinucleotide substitution types is modeled in the sample where mutations are detected, and a sample-specific substitution-specific support score threshold is set. In the hotspot detector, simpler heuristic rules are applied to detect variants listed at the target locus (typically genomic mutations with known clinical impact on cancer).

[0094] Support scores for each variant were calculated using a weighted linear combination of single-stranded, consensus, and double-stranded reads. For non-hotspot variants, detection was performed in two phases: (1) a core linear model and (2) an extended linear model. Based on the extended linear model, adaptive variant detection based on the core linear model was further filtered according to the probability scores. The extended linear model is a logistic regression classification model that uses two molecular counts and various techniques and biological variant features, such as sequence context, distance to read ends, strand bias, and the relative diffusion of variant families relative to the background.

[0095] (3) Filter and labeling module 606, which filters potential candidate SNV variants based on several cases not covered by the core linear model, such as double-strand ratio and optimization of the block list filtering, which are systematic experimental artifacts repeatedly seen in healthy samples.

[0096] Referring to Figure 6, the barcode copying module 600 is the same as described above with respect to Figure 1 for the InDel detector. Furthermore, apart from calculating different features for the SNV detector module, the preprocessing module 602 is also largely the same as described above with respect to Figure 1 for the InDel detector. The background correction steps are the same.

[0097] SNV detection module 604 includes essentially the same hotspot module as described above for the InDel detector, plus some additional rules applicable to SNV type mutations. For example, in some embodiments, an additional optional specific heuristic rule may include the special case for C>T or G>A substitutions, where a single duplex is required to support the read for sites identified as high-noise based on the background error distribution performed in the background correction step, or optionally, two duplexes are required to support the read.

[0098] The core linear model of the SNV detection module 604 is also basically the same as that described above for the InDel detector.

[0099] The trinucleotide module models the error distribution for each of the 12 substitution types or 192 trinucleotide substitution types in samples where mutations are detected, and sets sample-specific substitution-specific support score thresholds. For each ref>var substitution type (e.g., A>T), a weighted linear score of the variant support depth is collected for positions where AF < 0.5%. It is assumed that the vast majority of these positions are noise.

[0100] The result is the cumulative distribution of each substitution type, where position The value at this position indicates that it is greater than or equal to The number of replacements was observed at the support depth.

[0101] The minimum support threshold (weighted linear score) for each alternative type is determined individually. This minimum support threshold is used as the initial threshold and is adjusted by other factors such as gene weight and depth weight.

[0102] Extended Linear Model: The extended linear model can be a logistic regression model used for binary classification, which will be used to score the variants. The variant score (VS) is based on the count of single-stranded (s), consensus (c), and double-stranded (d) barcode families, as well as various additional technical and biological variant characteristics. These sequences are listed in Table 2 below.

[0103] Table 2

[0104]

[0105] The final variance score of the k additional features (f) and weights (w) is:

[0106]

[0107] in Represents the natural logarithm, and The probability that the value is between 0 and 1. Each feature coefficients or weights The model will learn from the training set containing known true positive and false positive variants. The implementation requires calculating additional metrics corresponding to these features using the output files from the barcode deduplication tool. Table 2 above describes the complete list of features in the extended linear model.

[0108] Adaptive detection of variants and two-stage scoring: First, a sample-specific adaptive model based on a linear model is used to generate preliminary detections. Here, a linearly weighted count of variant molecule counts is used. Then, an extended linear model is applied to these detections to filter out false positives and improve specificity. The principle behind using the two-stage model is illustrated in Figure 7.

[0109] In the two-stage approach, filtering based on extended linear scores significantly improves specificity with a relatively small decrease in sensitivity compared to using only the core linear model. In our tests, the one-stage scoring based on extended linear model scores did not perform well in practice. Features such as median variation distance, relative MAD, and chain bias depend on the number of molecules / families supporting the variant. When the variant molecule count is low, additional variant features may randomly take extreme values ​​and cannot be reliably estimated. The core linear model in the adaptive stage filters out low-molecule-count variants, thereby improving the accuracy of estimating additional features of scored variants in the second stage.

[0110] The scoring function for the second stage is further modified to filter out variants that take extreme values ​​for certain variability features. In these cases, the feature values ​​are transformed so that the scores become closer to 0. The following transformation is used:

[0111] 1. If disttrans < 20, then disttrans = -100000000

[0112] 2. If relvdistmad < 0.05, then relvdistmad = -100000000

[0113] 3. If relvstartmad < 0.05, then relvstartmad = -100000000

[0114] 4. If relvendmad < 0.05, then relvendmad = -100000000

[0115] 5. If relvstartmad < 0.15 & relvendmad < 0.15, then relvstartmad = -100000000, relvendmad = -100000000

[0116] 6. If relvstartmad < 0.10 & relvdistmad < 0.15, then relvstartmad = -100000000, relvdistmad = -100000000

[0117] 7. If relvdistmad < 0.15 & relvendmad < 0.10, then relvdistmad = -100000000, relvendmad = -100000000

[0118] These transitions are driven by observations from empirical data that include both true positive and true negative sites.

[0119] Referring to Figure 6, after the SNV detection module 604 generates a list of potential variants, additional filtering can be performed using the filtering module 606. The filtering module 606 may include a duplex ratio filter, a block list filter, and other tags that can be used to exclude certain variants from detection.

[0120] Double-strand ratio filter: For each variant detected, filter module 1006 calculates the number of deduplicated barcode reads supporting the variant and the reference allele (i.e., the number of barcode families). Using this barcode scheme, barcode families with double-strand support can also be identified (meaning the other strand in the original double-strand can also be assigned). This gives the count of deduplicated double-strand reads supporting the variant and the reference allele (i.e., the number of barcode families with double-strand evidence). The percentage of double-strand families should be around 10%-20%, so a significant depletion of double-strand reads that contradicts this expectation may indicate contamination of single-stranded molecules and artificial variants. Therefore, a statistic is defined from Fisher's exact test, and multiple test corrections are performed (for all variants across the sample). Significant variants are removed.

[0121] Block list filter. The block list filter can be generated in a manner similar to that described above for the InDel detector, and can exclude mutations with AF greater than 10%.

[0122] Tags: The tags that can be applied are basically the same as those described above for the InDel detector.

[0123] Exemplary System

[0124] Figure 8 illustrates a sequencing system 800 according to an embodiment of the present invention. The system shown includes a sample 805, such as an Xpandomer within a assay device 810, wherein the sample 805 can be assayed 808. For example, the sample 805 can be contacted with reagents of assay 808 to provide a signal of physical characteristic 815 (e.g., intensity signal) (e.g., sequence information of cell-free nucleic acid molecules). Assay 808 may include sequencing while expansion using assay device 810 (such as a nanopore sequencing device discussed above). Detector 820 detects physical characteristics 815 from the sample (e.g., fluorescence intensity, voltage, or current). Detector 820 may perform measurements at intervals (e.g., periodic intervals) to obtain data points constituting a data signal. In one embodiment, an analog-to-digital converter converts the analog signal from the detector into digital form multiple times.

[0125] The assay device 810 and detector 820 can form an assay system, such as a sequencing system 800 for sequencing according to embodiments described herein. Data signal 825 is transmitted from detector 820 to logic system 830. As an example, data signal 825 can be used to determine the sequence and / or location of nucleic acid molecules (e.g., DNA and / or RNA) in a reference genome. Data signal 825 can include various measurements performed simultaneously, such as different colored fluorescent dyes or different electrical signals of different molecules in sample 805, and therefore data signal 825 can correspond to multiple signals. Data signal 825 can be stored in local memory 835, external memory 840, or storage device 845. Sequencing system 800 can be composed of multiple assay devices 810 and detectors 820.

[0126] The logic system 830 may be or may include a computer system, an ASIC, a processor, a graphics processing unit (GPU), etc. It may also include or be coupled to a display (e.g., a monitor, LED display, etc.) and a user input device (e.g., a mouse, keyboard, buttons, etc.). The logic system 830 and other components may be part of a stand-alone or network-connected computer system, or they may be directly attached to or incorporated into a device including detector 820 and / or assay device 810 (e.g., a sequencing device). The logic system 830 may also include software executing in processor 850. The logic system 830 may include a computer-readable medium storing instructions for controlling the sequencing system 800 to perform any of the methods described herein. For example, the logic system 830 may provide commands to a system including assay device 810 to cause sequencing or other physical operations to be performed. Such physical operations may be performed in a specific order, e.g., reagents are added and removed in a specific order. Such physical operations can be performed by a robotic system (e.g., including a robotic arm), such as for obtaining samples and performing determination 808. The logic system 830 can also perform any steps of the computer processing methods described herein, such as, but not limited to, base detection, alignment, variant detection, the InDel detector algorithm shown in Figure 1, the barcode deduplication algorithm shown in Figure 2, and / or the SNV detector shown in Figure 6.

[0127] The sequencing system 800 may also include a treatment device 860, which can provide treatment to a subject. The treatment device 860 can determine the treatment and / or be used to administer the treatment. Examples of such treatments may include surgery, radiation therapy, chemotherapy, immunotherapy, targeted therapy, hormone therapy, and stem cell transplantation. A logic system 830 may be connected to the treatment device 860, for example, to provide the results of the methods described herein. The treatment device may receive input from other devices, such as imaging devices, and user input (e.g., to control the treatment, such as control of a robotic system).

[0128] The sequencing system 800 may also include a reporting device 855, which can present the results of any of the methods described herein, such as results determined using the sequencing system 800. The reporting device 855 can communicate with a reporting module within a logic system 830, which can summarize, format, and send reports to the reporting device 855. The reporting module can present information determined using any of the methods described herein. The information can be presented by the reporting device 855 in any format that can be recognized and interpreted by a user of the sequencing system 800. For example, the information can be presented by the reporting device 855 in a display, print, or transmission format, or any combination thereof.

[0129] Any computer system mentioned herein can utilize any suitable number of subsystems. Figure 9 shows an example of such a subsystem in computer system 900. In some embodiments, computer system 900 includes a single computer device, wherein the subsystem may be a component of said computer device. In other embodiments, computer system 900 may include multiple computer devices, each being a subsystem with internal components. The computer system may include desktop and laptop computers, tablets, mobile phones, and other mobile devices.

[0130] The subsystems shown in Figure 9 are interconnected via system bus 975. Additional subsystems are shown, such as printer 974, keyboard 978, storage devices 979, 982, monitor 976 (e.g., display screen, such as an LED, coupled to display adapter 982), etc. Peripheral devices and input / output (I / O) devices connected to I / O controller 971 can be connected via any number of components known in the art, such as input / output (I / O) ports 977 (e.g., USB, FireWire). ® The system 900 is connected to a computer system. For example, I / O port 977 or external interface 981 (e.g., Ethernet, Wi-Fi, etc.) can be used to connect computer system 900 to a wide area network, such as the Internet, a mouse input device, or a scanner. Interconnection via system bus 975 allows central processing unit 973 to communicate with each subsystem and control the execution of multiple instructions from system memory 972 or storage device 979 (e.g., a fixed disk, such as a hard disk drive, or optical disk), as well as the exchange of information between subsystems. System memory 972 and / or storage device 979 may contain computer-readable media. Another subsystem is a data collection device 985, such as a camera, microphone, accelerometer, etc. Any data mentioned herein can be output from one component to another and can be output to the user.

[0131] A computer system may include multiple identical components or subsystems, for example, connected together via an external interface 981, an internal interface, or a removable storage device that can be connected to or moved from one component to another. In some embodiments, the computer system, subsystem, or device may communicate via a network. In this case, one computer may be considered a client and another computer may be considered a server, with each computer being considered part of the same computer system. The client and server may each include multiple systems, subsystems, or components. In various embodiments, the method may involve a variety of numbers of clients and / or servers, including at least 10, 20, 50, 100, 200, 500, 1,000, or 10,000 devices. The method may include a variety of numbers of communication messages between devices, including at least 100, 200, 500, 1,000, 10,000, 50,000, 100,000, 500,000, or one million communication messages. Such communications may involve at least 1 MB, 10 MB, 100 MB, 1 GB, 10 GB, or 100 GB of data.

[0132] Various aspects of the embodiments can be implemented in a modular or integrated manner as control logic using hardware circuitry (e.g., application-specific integrated circuits or field-programmable gate arrays) and / or computer software stored in memory having a generally programmable processor. Therefore, the processor may include memory storing software instructions for configuring the hardware circuitry, and an FPGA or ASIC having configuration instructions. As used herein, the processor may include a single-core processor, a multi-core processor on the same integrated chip, or multiple processing units on a single circuit board or networked, as well as dedicated hardware. Based on the disclosure and teachings provided herein, those skilled in the art will recognize and understand other ways and / or methods of implementing the embodiments of this disclosure using hardware and combinations of hardware and software.

[0133] Any suitable computer language, such as Java, C, C++, C#, Objective-C, Swift, or scripting languages ​​such as Perl or Python, can be used, employing techniques such as conventional or object-oriented methods, to implement any software component or function described in this application as software code to be executed by a processor. The software code can be stored as a series of instructions or commands on a computer-readable medium for storage and / or transmission. Suitable non-transitory computer-readable media may include random access memory (RAM), read-only memory (ROM), magnetic media such as hard disk drives or floppy disks, or optical media such as optical discs (CDs) or DVDs (Digital Universal Optical Discs) or Blu-ray discs, flash memory, etc. The computer-readable medium can be any combination of such means. Furthermore, the order of operations can be rearranged. A process may terminate upon completion of its operations, but may have additional steps not included in the diagram. A process may correspond to a method, a function, a program, a subroutine, a subroutine, etc. When a process corresponds to a function, its termination may correspond to the function returning to a detection function or the main function.

[0134] Such programs can also be encoded and transmitted using carrier signals adapted for transmission over wired, optical, and / or wireless networks conforming to various protocols, including the Internet. Thus, computer-readable media can be created using data signals encoded with such programs. Computer-readable media encoded with program code can be packaged with compatible devices (e.g., as firmware) or provided separately from other devices (e.g., for download via the Internet). Any such computer-readable media can reside on or within a single computer product (e.g., a hard drive, CD, or an entire computer system) and can exist on or within different computer products within a system or network. A computer system may include a monitor, printer, or other suitable display for providing the user with any of the results mentioned herein.

[0135] Any method described herein can be performed wholly or partially using a computer system comprising one or more processors, which can be configured to perform the steps. Any operation performed using a processor can be performed in real time. The term "real time" can refer to a computational operation or process completed within a certain time limit. For example, the time limit could be 30 seconds, 1 minute, 10 minutes, 30 minutes, 1 hour, 4 hours, 1 day, or 7 days. Therefore, embodiments may be directed to a computer system configured to perform the steps of any method described herein, possibly having different components performing the corresponding steps or groups of corresponding steps. Although the steps are presented as numbered, the steps of the methods described herein may be performed simultaneously or at different times or in different orders. Furthermore, some steps may be used in conjunction with some steps from other methods. Additionally, all or some steps may be optional. Furthermore, any step of any method can be performed using modules, units, circuits, or other means of a system for performing these steps.

[0136] The specific details of a particular embodiment may be combined in any suitable manner without departing from the spirit and scope of the embodiments disclosed herein. However, other embodiments of this disclosure may be directed to specific embodiments relating to each individual aspect, or specific combinations of these individual aspects.

[0137] For purposes of illustration and description, the foregoing description of exemplary embodiments of the present disclosure has been presented. It is not intended to be exhaustive or to limit the disclosure to the precise forms described, and many modifications and variations are possible in accordance with the foregoing teachings.

[0138] When a feature or element is referred to herein as being “on” another feature or element, it may be directly on the other feature or element, or there may be intermediate features and / or elements present. Conversely, when a feature or element is referred to as being “directly on” another feature or element, there are no intermediate features or elements present. It will also be understood that when a feature or element is referred to as being “connected,” “attached,” or “coupled” to another feature or element, it may be directly connected, attached, or coupled to the other feature or element, or there may be intermediate features or elements present. Conversely, when a feature or element is referred to as being “directly connected,” “directly attached,” or “directly coupled” to another feature or element, there are no intermediate features or elements present. Although one embodiment has been described or illustrated, the features and elements thus described or illustrated may be applied to other embodiments. Those skilled in the art will also recognize that a structure or feature referred to as being “adjacent” to another feature may have portions that overlap with or are located below the adjacent feature.

[0139] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. For example, as used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “including” are used in this specification, they specify the presence of the defined features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof. As used herein, the term “and / or” includes any and all combinations of one or more of the associated listed items and may be abbreviated to “ / ”.

[0140] For ease of description, spatially relative terms such as “below,” “below,” “lower than,” “above,” “over,” etc., are used herein to describe the relationship of one element or feature to another, as illustrated in the accompanying drawings. It should be understood that, in addition to the orientations depicted in the drawings, spatially relative terms are also intended to cover different orientations of the device in use or operation. For example, if the device in the drawings is inverted, an element described as “below” or “under” other elements or features would then be oriented “above” other elements or features. Thus, the exemplary term “below” can encompass both the orientations of “above” and “below.” The device may be oriented in other ways (rotated 90 degrees or in other orientations), and the spatially relative descriptive terms used herein shall be interpreted accordingly. Similarly, unless otherwise specifically indicated, terms such as “up,” “down,” “vertical,” “horizontal,” etc., are used herein for illustrative purposes only.

[0141] Although the terms “first” and “second” may be used herein to describe various features / elements (including steps), these features / elements should not be limited by these terms unless the context otherwise indicates. These terms can be used to distinguish one feature / element from another. Therefore, without departing from the teachings of the invention, the first feature / element discussed below may be referred to as the second feature / element, and similarly, the second feature / element discussed below may be referred to as the first feature / element.

[0142] Throughout this specification and the following claims, unless the context otherwise requires, the word “comprising” and variations such as “including” and “containing” mean that various components may be used together in methods and articles (e.g., compositions and devices that include means and methods). For example, the term “comprising” will be understood to imply the inclusion of any of the specified elements or steps, but does not exclude any other elements or steps.

[0143] As used herein in the specification and claims, including as in the examples, and unless otherwise expressly specified, all figures may be interpreted as if preceded by the words “about” or “approximately,” even if the term is not explicitly stated. When describing magnitude and / or location, the phrase “about” or “approximately” may be used to indicate that the described value and / or location is within a reasonably expected range of value and / or location. For example, numerical values ​​may have values ​​of + / - 0.1% of a specified value (or range of values), + / - 1% of a specified value (or range of values), + / - 2% of a specified value (or range of values), + / - 5% of a specified value (or range of values), + / - 10% of a specified value (or range of values), etc. Unless the context otherwise indicates, any numerical value given herein should also be understood to include about or approximately that value. For example, if the value “10” is disclosed, “about 10” is also disclosed. Any numerical range described herein is intended to include all subranges contained therein. It should also be understood that, as those skilled in the art would appropriately understand, when a value is disclosed, the terms "less than or equal to" that value, "greater than or equal to" that value, and possible ranges between values ​​are also disclosed. For example, if the value "X" is disclosed, "less than or equal to X" and "greater than or equal to X" are also disclosed (e.g., in the case where X is a numerical value). It should also be understood that throughout this application, data is provided in a variety of different formats, and the data represents a range of endpoints and start points, as well as any combination of data points. For example, if specific data point "10" and specific data point "15" are disclosed, it should be understood that values ​​greater than, greater than or equal to, less than, less than or equal to, equal to 10 and 15, and values ​​between 10 and 15 are considered disclosed. It should also be understood that each unit between two specific units is also disclosed. For example, if 10 and 15 are disclosed, 11, 12, 13, and 14 are also disclosed.

[0144] Although various illustrative embodiments have been described above, any of a variety of changes may be made to the various embodiments without departing from the scope of the invention as described in the claims. For example, in alternative embodiments, the order in which the various method steps described are performed may often be changed, while in other alternative embodiments, one or more method steps may be skipped entirely. In some embodiments, optional features of various apparatus and system embodiments may be included, while in other embodiments they may not be included. Therefore, the foregoing description is provided primarily for illustrative purposes and should not be construed as limiting the scope of the invention as set forth in the claims.

[0145] The examples and illustrations included herein are shown in an illustrative and not limiting manner, manner, of specific embodiments in which the subject matter may be practiced. As mentioned, other embodiments may be utilized and derived therefrom, allowing for structural and logical substitutions and changes without departing from the scope of this disclosure. These embodiments of the subject matter of the invention may be referred to herein individually or collectively by the term "invention," merely for convenience and not intended to actively limit the scope of this application to any single invention or inventive concept, given that more than one invention or inventive concept has actually been disclosed. Therefore, although specific embodiments have been shown and described herein, any arrangement calculated to achieve the same purpose may replace the specific embodiments shown. This disclosure is intended to cover any and all modifications or variations of the various embodiments. After reviewing the foregoing description, combinations of the above embodiments and other embodiments not explicitly described herein will be apparent to those skilled in the art.

Claims

1. A method for detecting insertions and deletions (InDels) from sequencing data generated from multiple samples, the method comprising: Receive a computer file containing multiple paired sequence read segments; The multiple paired sequence reads are compared with a reference sequence; Sort the aligned and paired sequence reads; Based on the unique molecular identifier (UMI), the start position of the sequence read, and the end position of the sequence read, the sorted, aligned, and paired sequence reads are grouped into multiple groups, where each member of a group has the same UMI, the start position of the sequence read, and the end position of the sequence read; Determine the consensus sequence from the members of each group; Parse the paired consensus sequence reads; Correct the parsed consensus sequence reads; A set of candidate InDel variants is determined by comparing the corrected consensus sequence reads with the reference sequence; For each candidate InDel variant, a weighted linear score is calculated using a linear model, wherein the candidate InDel variant is identified as a potential InDel variant when the weighted linear score is greater than a linear model threshold. as well as For each potential InDel variant, a score is calculated using a non-linear tree-based ensemble machine learning classifier, where the potential InDel variant is identified as a true InDel variant when the score is greater than a threshold set by the machine learning classifier.

2. The method of claim 1, further comprising generating a computer file containing the paired consensus sequence reads.

3. The method of claim 1, wherein the step of correcting the parsed consensus sequence reads comprises: The parsed consensus sequence reads are compared with the reference sequence to determine a set of preliminary InDel variants, including a count of each preliminary InDel variant; as well as The count of each preliminary InDel variant is adjusted based on a comparison of the count of each preliminary InDel variant with the corresponding background variant count determined from the control sample.

4. The method of claim 3, wherein the comparison includes determining whether the count of each preliminary InDel variant is background noise.

5. The method of claim 4, wherein the step of determining whether the count of each preliminary InDel variant is background noise includes calculating the probability that the preliminary InDel variant is background noise.

6. The method according to claim 1, wherein the linear model includes double-chain mutation count, single-chain mutation count, and consensus mutation count.

7. The method of claim 1, wherein the threshold of the linear model is sample-specific.

8. The method of claim 7, wherein for all the candidate InDel variants in the sample, a threshold for the linear model is determined by performing a regression fit on a graph of the cumulative candidate InDel variant counts as a function of the weighted linear score, and setting the threshold at a weighted linear score that produces a target cumulative InDel variant count, the cumulative candidate InDel variant count having at least some weighted linear score.

9. The method of claim 1, wherein the threshold of the linear model is determined from a set of training data by: determining the cumulative count of false positive variant detections as a function of a weighted linear score, and selecting the weighted linear score corresponding to a target cumulative false positive count as the threshold.

10. The method of claim 1, wherein the nonlinear tree-based ensemble machine learning classifier comprises single-stranded molecule counts, consensus molecule counts, double-stranded molecule counts, the ratio of the single-stranded molecule counts to the double-stranded molecule counts, the median distance of the latent variant from the 5' end of the sequence read, the relative diffusion of variant distances in variant molecules relative to the background, the relative diffusion of insertion start positions in variant molecules relative to the background, the original alignment chain bias of consensus molecules, the original origin chain bias of single-stranded molecules, the original alignment chain bias of non-double-stranded molecules, the proportion of double-stranded variant molecules with two reads, the proportion of double-stranded variant molecules with less than or equal to three reads, the proportion of consensus variant molecules with two reads, the proportion of consensus variant molecules with less than or equal to three reads, the relative frequency of single-stranded molecules in variant molecules relative to the background, the relative frequency of double-stranded molecules in variant molecules relative to the background, and the 2 x 2 Fisher exact test p-value of the ratio of single-stranded molecules to double-stranded molecules in variant molecules relative to the background.

11. The method of claim 1, wherein a set of healthy samples is used to calibrate the threshold of the machine learning classifier to achieve a predetermined false positive rate.

12. A method for detecting single nucleotide variants (SNVs) from sequencing data generated from multiple samples, the method comprising: Receive a computer file containing multiple paired sequence read segments; The sequence read is compared with a reference sequence; The aligned sequence reads are sorted according to their pairs; The sorted, aligned and paired sequence reads are grouped into multiple groups using a unique molecular identifier (UMI), the start position of the sequence read, and the end position of the sequence read. Each member of one group has the same UMI, the start position of the sequence read, and the end position of the sequence read. Determine the consensus sequence from the members of each group; Parse the paired consensus sequence reads; Correct the parsed consensus sequence reads; A set of candidate variants was determined based on a comparison of corrected consensus reads; For each candidate SNV variant, a weighted linear score is calculated using a linear model, wherein a candidate SNV variant is identified as a potential SNV variant when the weighted linear score is greater than a linear model threshold; and For each potential SNV variant, a weighted logistic score is calculated using a logistic regression model, where the potential SNV variant is identified as a true variant when the weighted logistic score is greater than a threshold of the logistic regression model.

13. The method of claim 12, the method further comprising generating a computer file containing paired consensus sequence read segments.

14. The method of claim 12, wherein the step of correcting the parsed consensus sequence reads comprises: The parsed consensus read is compared with the reference sequence to determine a set of preliminary SNV variants, including a count of each preliminary SNV variant; as well as The count of each initial SNV variant is adjusted based on a comparison of the count of each initial SNV variant with the corresponding background variant count determined from the control sample.

15. The method of claim 14, wherein the comparison includes determining whether the count of each preliminary SNV variation is background noise.

16. The method of claim 15, wherein the step of determining whether the count of each preliminary SNV variation is background noise includes calculating the probability that the preliminary SNV variation is background noise.

17. The method of claim 12, wherein the linear model comprises a double-chain mutation count, a single-chain mutation count, and a consensus mutation count.

18. The method of claim 12, wherein the threshold of the linear model is sample-specific.

19. The method of claim 18, wherein for all the candidate SNV variants in the sample, a threshold for the linear model is determined by performing a regression fit on a graph of the cumulative candidate SNV variant counts as a function of the weighted linear score, and setting the threshold at a weighted linear score that produces a target cumulative SNV variant count, the cumulative candidate SNV variant count having at least some weighted linear score.

20. The method of claim 12, wherein the threshold of the linear model is determined from a set of training data by: determining the cumulative count of false positive variant detections as a function of a weighted linear score, and selecting the weighted linear score corresponding to a target cumulative false positive count as the threshold.

21. The method according to claim 12, wherein the logistic regression model includes a double-chain mutation count, a single-chain mutation count, and a consensus mutation count.

22. The method of claim 21, wherein the logistic regression model further comprises the ratio of the single-stranded variant count to the double-stranded molecule count, adjustments for coverage depth, adjustments for errors in the gene, the median distance of the variant from the 5' end of the sequence read, the relative diffusion of variant distance in the variant molecule relative to the background, the relative diffusion of insertion start position in the variant molecule relative to the background, the relative diffusion of insertion end position in the variant molecule relative to the background, the original alignment strand bias, the original origin strand bias, the substitution type of the variant, the two base pairs surrounding the variant position, the proportion of double-stranded variant molecules with two reads, the proportion of double-stranded variant molecules with less than or equal to three reads, the proportion of consensus variant molecules with two reads, the proportion of consensus variant molecules with less than or equal to three reads, the relative frequency of consensus in the variant molecule relative to the background, and the relative frequency of double strands in the variant molecule relative to the background.

23. The method of claim 12, wherein the logistic regression model comprises a plurality of features, wherein each feature is multiplied by a weight, wherein the weights are determined from a set of training data having known true positive variants and known false positive variants.

24. A system comprising: Measuring device; and A logic system comprising at least a processor and a memory storing a set of instructions, wherein the processor, when executing the instructions, is configured to perform the method according to any one of claims 1 to 11.

25. The system of claim 24, further comprising a treatment device for determining treatment or administering treatment to the patient based on at least one true InDel variant.

26. The system of claim 24, further comprising a reporting device for displaying information about at least one genuine InDel variant.

27. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform the method according to any one of claims 1 to 11.

28. A system comprising: Measuring device; and A logic system comprising at least a processor and a memory storing a set of instructions, wherein the processor, when executing the instructions, is configured to perform the method according to any one of claims 12 to 23.

29. The system of claim 28, further comprising a treatment device for determining treatment or administering treatment to the patient based on at least one real variant.

30. The system of claim 28, further comprising a reporting device for displaying information about at least one real variation.

31. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform the method according to any one of claims 12 to 23.

Citation Information

Patent Citations

  • Nanopore Based Molecular Detection and Sequencing

    US20130244340A1

  • DNA sequencing by synthesis using modified nucleotides and nanopore detection

    US20130264207A1

  • Nucleic acid sequencing using tags

    US20140134616A1

  • Nucleic acid sequencing by nanopore detection of tag molecules

    US20150119259A1

  • Methods for creating bilayers for use with nanopore sensors

    US20150337366A1