A method for constructing a background noise filtering model for repeated regions and a background noise filtering method

Abstract

The present invention relates to the technical field of gene mutation detection, and discloses a method for constructing a background noise filtering model for a repeat region and a background noise filtering method. For point mutations and insertion and deletion mutations, a background baseline is constructed according to negative samples of a panel, a repeat region database based on the human genome is constructed, and finally, based on the abundance of SNV and indel mutations, the background baseline, the relationship with the repeat region and other information, it is judged whether the mutation is background noise, and then classified and filtered. The method for filtering sequencing background noise of the present invention has shown excellent accuracy and efficiency in the identification and application of false positive mutations and true positive mutations.

Description

Technical Field

[0001] The present invention relates to the technical field of gene mutation detection, and in particular to a method for constructing a repeated region background noise filtering model and a background noise filtering method. Background Art

[0002] Point mutations (single nucleotide mutations, SNVs) and insertion-deletion mutations (InDels) are common forms of variation in the human genome. Many important point mutations and insertion-deletion mutations are carcinogenic driver mutations, and there are more and more opportunities for targeted therapy. There are also more and more sequencing platforms and mutation identification algorithms for point mutations and insertion-deletion mutations. Common methods for detecting gene point mutations include PCR, Sanger sequencing (first-generation sequencing), and second-generation sequencing (NGS).

[0003] The second generation sequencing has the characteristics of high throughput, and the sequencing cost is also decreasing year by year. However, due to the current technical limitations of the second generation sequencing, false positive results often appear in the final analysis results, including false positive noise caused by a large number of sequencing platforms, especially in the simple repetitive sequence regions on the genome, which are more prone to false positive background noise. At present, the commonly used tools for filtering point mutations and insertion and deletion mutations include GATK Variant Filtration, Varscan fp filter, etc., but these tools are lacking in the accuracy of removing sequencing background noise, especially for false positives in repetitive regions, the filtering effectiveness is poor. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for constructing a background noise filtering model for a repeat region and a background noise filtering method for point mutations and insertion / deletion mutations, in order to solve the technical problem of poor effectiveness in filtering false positive noise in the existing second-generation sequencing technology for point mutations and insertion / deletion mutations.

[0005] In order to solve the above technical problems, in a first aspect, the present invention provides a method for constructing a repeated region background noise filtering model, comprising the following steps:

[0006] (1) performing next-generation sequencing on the sample and obtaining the SNV and indel mutations of the sample by comparing it with the reference genome;

[0007] (2) extracting low-complexity regions and a simple repeat region list from the reference genome repeat region list; identifying simple repeat sequences on the reference genome; merging the simple repeat sequence region with the repeat region list according to base positions to obtain a repeat region list; and determining a training set based on the repeat region list;

[0008] (3) constructing a random forest model for identifying true positive and false positive mutations using the training set according to the characteristic values, thereby completing the model construction;

[0009] The characteristic value includes at least one of the following values:

[0010] the length of the mutation;

[0011] AF;

[0012] ratio.sum value;

[0013] The number of all paired-end reads supporting the mutation;

[0014] The number of all reads supporting the mutation;

[0015] Among them, AF is the actual mutation abundance of the mutation to be filtered; the ratio.sum value is the ratio of AF to the average abundance of the mutation in the background baseline; the average abundance is the ratio of AD.sum to DP.sum; the AD.sum value is the sum of the number of sequencing sequences that support the mutation in the sample; and the DP.sum value is the sum of the number of sequencing sequences that cover the mutation.

[0016] As a preferred implementation of the method for constructing a background noise filtering model for a repeat region of the present invention, the training set is the repeat region list, each repeat region is extended 10bp upstream and downstream, and true positive mutations and false positive mutations on exons of similar numbers are specified within this range.

[0017] As a preferred embodiment of the method for constructing a background noise filtering model for a repeat region of the present invention, the simple repeat region refers to a repeat region with a repeat unit ≤ 6 bp.

[0018] As a preferred implementation of the method for constructing a repeated region background noise filtering model of the present invention, in step (1), the samples are negative samples, and the number of samples is not less than 40.

[0019] As a preferred embodiment of the method for constructing a background noise filtering model for a repetitive region of the present invention, in the step (1), after obtaining the snv and indel mutations of the sample, the DP.sum value, AD.sum value, DP.max value, and AD.max value of the sample are counted;

[0020] Among them, the DP.max value is the maximum number of sequencing sequences in the sample that cover the mutation; the AD.max value is the maximum number of sequencing sequences in the sample that support the mutation.

[0021] As a preferred implementation of the method for constructing a repeated region background noise filtering model of the present invention, in the step (1), a filtered background baseline is constructed; the evaluation value of the background baseline includes at least one of the following:

[0022] ratio.sum value;

[0023] ratio.max value; the ratio.max value is the ratio of the maximum abundance of the mutation in AF to the background baseline; the maximum abundance is the ratio of AD.max to DP.max;

[0024] pvalue.sum value; the pvalue.sum value is the Fisher one-sided test p-value value of AD, DP, DP.sum, and AD.sum;

[0025] Among them, the AD is the number of sequencing sequences that actually support mutations of the mutation to be filtered; and the DP is the number of sequencing sequences that actually cover mutations of the mutation to be filtered.

[0026] Preferably, when a mutation does not exist in the background baseline, the DP.sum and DP.max are given a virtual value of 100000, and the AD.sum and AD.max are given a virtual value of 1 for calculation.

[0027] As a preferred embodiment of the method for constructing a background noise filtering model for a repeat region of the present invention, in the step (2), a method for annotating the repeat region of an intron region is included:

[0028] According to the specified transcript, the Exon position corresponding to the transcript is found on GENECODE, and the merged repeat region list is annotated.

[0029] As a preferred embodiment of the method for constructing a background noise filtering model for a repeat region of the present invention, in step (2), the repeat region in the intron region is a simple repeat region in which the start and end base positions of the repeat region are on the intron and are more than 2 bp away from the exon, and the minimum repeat unit is repeated 5 times or more in series.

[0030] As a preferred implementation of the method for constructing a background noise filtering model for a repeated region of the present invention, in the step (3), when the mutation is a snv mutation, the length of the mutation is equal to 0.

[0031] As a preferred implementation of the method for constructing a repeated region background noise filtering model of the present invention, in the step (3), the characteristic value further includes:

[0032] The minimum distance between the mutated base position and the nearest repeat region; or

[0033] The indel base matches the nearest repeated base in the repeated region, and the calculation formula of the matching score is: score / length.

[0034] Preferably, the minimum distance between the mutated base position and the nearest repeat region ranges from 0 to 10. If the mutated base position is within the repeat region, the minimum distance is 0.

[0035] In a second aspect, the present invention provides a device comprising the repetitive region background noise filtering model constructed by the method.

[0036] In a third aspect, the present invention provides a method for filtering sequencing background noise, comprising:

[0037] Identify true positive and false positive mutations of the sample, and filter out false positive mutations;

[0038] The filtering uses a characteristic value, and the characteristic value includes at least one of the following:

[0039] The number of bases in the nearest repeat region; the length of the indel bases; the matching score between the indel bases and the repeat region; the ratio.sum value; the number of all reads supporting the mutation; the number of all paired-end reads supporting the mutation; AF;

[0040] Among them, AF is the actual mutation abundance of the mutation to be filtered; the ratio.sum value is the ratio of AF to the average abundance of the mutation in the background baseline; the average abundance is the ratio of AD.sum to DP.sum; the AD.sum value is the sum of the number of sequencing sequences that support the mutation in the sample; and the DP.sum value is the sum of the number of sequencing sequences that cover the mutation.

[0041] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, it is characterized in that the filtering includes the following conditions:

[0042] 1) For deletion mutations, the reference base sequence of the mutation is required to overlap with the intron region repeat region by at least one base; for insertion mutations, the starting base position of the intron region repeat region is required to be ≤ the base position of the mutation on the chromosome ≤ the ending base position of the intron region repeat region, and / or the base position of the mutation on the chromosome is the previous position of the starting base position of the intron region repeat region; and

[0043] 2) The bases of the mutation indel are completely composed of tandem repeats of the minimum repeating unit of the repeat region, or the bases of the indel are part of the repeating unit.

[0044] In a fourth aspect, the present invention further provides a method for filtering sequencing background noise, comprising:

[0045] Filter out false positive mutations based on a specific threshold;

[0046] The characteristic value of the specific threshold includes at least one of the following: AF, ratio.sum value, pvalue.sum value;

[0047] Among them, AF is the actual mutation abundance of the mutation to be filtered; the ratio.sum value is the ratio of AF to the average abundance of the mutation in the background baseline; the average abundance is the ratio of AD.sum to DP.sum; the AD.sum value is the sum of the number of sequencing sequences supporting the mutation in the sample; the DP.sum value is the sum of the number of sequencing sequences covering the mutation; AD is the number of sequencing sequences that actually support the mutation to be filtered; and DP is the number of sequencing sequences that actually cover the mutation to be filtered;

[0048] The pvalue.sum value is the Fisher one-sided test p-value of the four values ​​of AD, DP, DP.sum, and AD.sum.

[0049] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, it is characterized in that the filtering further comprises the following conditions:

[0050] 1) For deletion mutations, the reference base sequence of the mutation is required to overlap with the intron region repeat region by at least one base; for insertion mutations, the starting base position of the intron region repeat region is required to be ≤ the base position of the mutation on the chromosome ≤ the ending base position of the intron region repeat region, and / or the base position of the mutation on the chromosome is the previous position of the starting base position of the intron region repeat region; and

[0051] 2) The bases of the mutation indel are completely composed of tandem repeats of the minimum repeating unit of the repeat region, or the bases of the indel are part of the repeating unit.

[0052] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, the characteristic value also includes:

[0053] The number of bases in the nearest repeat region; the length of the bases of the indel; the matching score between the bases of the indel and the repeat region; the number of all reads supporting the mutation; the number of all double-end reads supporting the mutation.

[0054] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, the calculation formula of the pvalue.sum value is as follows:

[0055]

[0056] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, when the sample is a tissue sample and the mutation type is SNV or indel, filtering is performed according to the following conditions to remove false positive mutations: for HOT sites, filtering is performed according to a specific threshold value AF≥1%; for non-HOT sites, filtering is performed according to a specific threshold value AF≥2%.

[0057] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, when the sample is a tissue sample, the mutation type is SNV or indel and there is a background baseline, filtering is performed according to the following conditions to remove false positive mutations: for HOT sites, filtering is performed according to a specific threshold ratio.sum≥5; for non-HOT sites, filtering is performed according to a specific threshold ratio.sum≥10.

[0058] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, the sample is a blood sample, and when the mutation type is SNV or indel, filtering is performed according to the following conditions to remove false positive mutations, and filtering is performed according to a specific threshold value pvalue.sum≤0.0001.

[0059] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, the sample is a blood sample. When the mutation type is SNV or indel, it is filtered according to the following conditions to remove false positive mutations: for HOT sites, it is filtered according to a specific threshold AF≥0.1%; for non-HOT sites, it is filtered according to the specific threshold AF≥0.2%.

[0060] As a preferred embodiment of the method for filtering sequencing background noise of the present invention, the sample is a blood sample, and when the mutation type is SNV or indel and there is a background baseline, it is filtered according to the following conditions to remove false positive mutations: for HOT sites, it is filtered according to a specific threshold ratio.sum≥5; for non-HOT sites, it is filtered according to a specific threshold ratio.sum≥10.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] For the repeat regions of point mutations and insertion-deletion mutations, the repeat region background noise filtering model of the present invention has high accuracy and effectiveness. The repeat region background noise filtering model construction method of the present invention is simple and easy to operate, which is conducive to promotion and use. The method of filtering sequencing background noise of the present invention has a good filtering effect, and can achieve a false positive filtering rate of 98.74% for tissue samples and 99.68% for blood samples; while the false positive filtering rate is improved, the false positive misjudgment rate is low, the accuracy is high, and the accuracy rate of true positive identification is above 99%.

[0063] In summary, for point mutations and insertion-deletion mutations, the method of filtering sequencing background noise of the present invention has shown excellent accuracy and efficiency in identifying false positive mutations and true positive mutations. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A diagram showing the relationship between the noise filtering steps and the files used in the embodiment;

[0065] Figure 2 This is the experimental flow chart for filtering the background noise of sequencing samples in the verification example;

[0066] Figure 3 This is the confusion matrix diagram of the blood sample in the validation example;

[0067] Figure 4 This is the confusion matrix diagram of the tissue samples in the validation example. DETAILED DESCRIPTION

[0068] Below in conjunction with the drawings and Examples, the specific embodiments of the present invention are described in further detail. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention. In the description of the present invention, it should be understood that the materials or methods involved, if not otherwise specified, are conventional materials or methods.

[0069] The abbreviations and terms used in this invention are as follows:

[0070] SNV is a point mutation (single-nucleotide variant);

[0071] Indel is an insertion-deletion mutation, deletion mutation is a deletion mutation, and insertion mutation is an insertion mutation;

[0072] Indel bases: Compared with the reference genome, the mutated sequence has inserted or deleted bases;

[0073] The mutation abundance is the number of reads supporting the mutation divided by the number of reads covering the mutation position, that is, the allele frequency of the mutation, expressed as AF or VAF;

[0074] BAM format files are binary files of SAM format files; SAM (Sequence Alignment / Mapformat) files are a sequence alignment format standard established by Sanger, and are text formats with TAB as delimiters;

[0075] Low_complexity is the low-complexity region of the sequence on the genome;

[0076] AD is the number of reads that actually support the mutation, that is, the number of sequencing reads that support the mutation;

[0077] DP is the actual mutation depth, that is, the number of all sequencing reads covering the mutation position.

[0078] Embodiment: A method for filtering sequencing background noise

[0079] For SNV and indel mutations, this paper builds a background baseline based on the negative samples of the panel, builds a database of repeated regions based on the human genome, and finally determines whether the mutation is background noise based on the abundance of SNV and indel mutations, background baseline, and relationship with repeated regions, and then classifies and filters them. The specific steps include:

[0080] 1. Construction of background baseline and calculation of related evaluation values

[0081] For different panels, at least 40 negative samples are used to construct the background baseline. For blood samples, white blood cell samples or negative blood samples of healthy people can be used as construction samples. For tissue samples, adjacent cancer tissue samples or negative tissue samples of healthy people can be used as construction samples.

[0082] The samtools software was used to analyze the BAM files aligned to the human genome to obtain the snv and indel mutation results, and the snv and indel mutations were counted separately: the sum of the number of sequencing sequences covering the mutation in all samples, i.e. DP.sum; the sum of the number of sequencing sequences supporting the mutation in all samples, i.e. AD.sum; the maximum number of sequencing sequences covering the mutation in all samples, i.e. DP.max; the maximum number of sequencing sequences supporting the mutation in all samples, i.e. AD.max.

[0083] The background baseline estimates used during filtering are:

[0084] (1) ratio.sum value: the ratio of the actual mutation abundance (AF) to the average abundance of the mutation in the background baseline (AD.sum / DP.sum); (2) ratio.max value: the ratio of the actual mutation abundance (AF) to the maximum abundance of the mutation in the background baseline (AD.max / DP.max); (3) pvalue.sum value: the Fisher one-sided test p-value of the four values ​​of the actual number of reads supporting the mutation (AD), the actual mutation depth (DP), DP.sum, and AD.sum. It should be noted that if a mutation does not exist in the background baseline, DP.sum and DP.max are given a virtual value of 100,000, and AD.sum and AD.max are given a virtual value of 1 for calculation.

[0085] 2. Construction and annotation of repeated regions

[0086] Extract the repeat regions identified by the tandem repeats finder software with low complexity and all repeat units ≤ 6 bp from the human genome repeat region list provided by the UCSC database (http: / / genome.ucsc.edu), or use RepeatMasker software to identify all repeat regions with low complexity and all repeat units <= 6 bp on the human genome. At the same time, use misa software to identify simple repeat sequences on the human genome, and merge the identified simple repeat sequence regions with the above repeat region list according to the base position.

[0087] Further, the repeat region list of the intron region is obtained through the repeat region list: according to the specified transcript, the Exon position and other information corresponding to the transcript are found on GENECODE, and the merged repeat region list is annotated. The repeat region list is classified as the repeat region list of the intron region, where the start and end base positions of the repeat region are on Intron, the distance from Exon is more than 2bp, and the minimum repeat unit is repeated 5 times or more in tandem.

[0088] 3. Construction of background noise filtering model for repeated regions

[0089] According to the 7 eigenvalues, a random forest model for identifying true positive and false positive mutations was constructed using the training set.

[0090] Training set construction: In the list of repeat regions constructed above, each repeat region is extended by 10 bp upstream and downstream, and true positive mutations and false positive mutations on exons with similar numbers are specified within this range to distinguish blood and tissue samples.

[0091] The characteristic values ​​include: (1) the length of the indel mutation, which is equal to 0 for the snv mutation; (2) the ratio.sum value calculated based on the background baseline; (3) the number of all paired-end reads that support the mutation (after PCR deduplication); (4) the number of all reads that support the mutation (after PCR deduplication); (5) the minimum distance between the mutated base position and the nearest repeat region (from the start and end positions of the repeat region), where the distance is 0 within the repeat region and the value range is 0 to 10; (6) the abundance of the mutation (AF or VAF); (7) the match score between the indel base and the repeat base of the nearest repeat region, calculated as: score / length, score: the base and the repeat unit of the repeat region, 1 point for a match, 0.2 points for a mismatch, and 0.1 points for a mismatch but the base is within the repeat base of the repeat region.

[0092] For example, as shown in Table 1, assume that a repeating region has a repeating unit of AC: ACACACACAC.

[0093] Table 1 Matching scores of repeated bases

[0094] mutation Matching score T->TAC (1+1) / 2 T->A 1 / 1 T->C 1 / 1 T->G -0.2 / 1 T->TACC (1+1-0.1) / 3 T->TACG (1+1-0.2) / 3

[0095] The above method was used to determine the training data sets of blood samples and tissue samples, and the seven eigenvalues ​​of each mutation in the data sets were calculated to construct random forest models for blood samples and tissue samples, respectively.

[0096] 4. Background noise filtering

[0097] The files used for filtering are as follows Figure 1 The specific steps are as follows:

[0098] In the first step (step 1), indel mutations that meet the following two conditions will be filtered out:

[0099] (1) For deletion mutations, the reference base sequence of the mutation is required to overlap with the intronic repeat region by at least one base; for insertion mutations, the starting base position of the intronic repeat region is required to be ≤ the base position of the mutation on the chromosome ≤ the ending base position of the intronic repeat region, or the base position of the mutation on the chromosome is before the starting base position of the intronic repeat region; for example: the 2488178-2488182 position of chromosome 1 is a repeat of a segment of C bases, the 2488177 position of chromosome 1 has a GC>G mutation, and there is a 1-base overlap in the repeat region, which meets the requirement; the 2488177 position of chromosome 1 has a G>GCC mutation, which is before the starting base position of the repeat region, which also meets the requirement.

[0100] (2) The bases inserted or deleted in the Indel mutation are completely composed of tandem repeats of the minimum repeating unit of the repeat region, or the bases inserted or deleted are part of the repeating unit. For example, in the TACCACCACCACCACC repeat region, T>TACC and T>TACCAC both meet the requirements, but T>TC does not.

[0101] In the second step (step 2), for mutations that are not filtered in the above step, if the mutation is located in the repeat region and within 10bp upstream and downstream (the starting base position of the repeat region -10bp ≤ the base position of the mutation on the chromosome ≤ the ending base position of the repeat region +10bp), the following 7 characteristic values ​​are calculated: the number of bases in the nearest repeat region, the length of the indel bases, the matching score of the indel bases with the repeat region, the ratio.sum value calculated based on the background baseline, the number of all reads supporting the mutation (after PCR deduplication), the number of all paired-end reads supporting the mutation (after PCR deduplication), the mutation abundance, and the false positive filtering model of the repeat region is used to identify true positive and false positive mutations and filter out false positive mutations.

[0102] In the third step (step three), mutations that are not filtered in the above steps are filtered according to specific thresholds using the ratio.sum value, ratio.max value, and pvalue.sum value calculated based on mutation abundance and background baseline.

[0103] Among them, the filtering thresholds of tissue samples are shown in Table 2. Mutations that meet the following conditions are considered to be true positive mutations, otherwise they are false positive mutations.

[0104] Table 2 Filtering thresholds for tissue samples

[0105]

[0106]

[0107] Among them, the filtering thresholds of blood samples are shown in Table 3. Mutations that meet the following conditions are considered to be true positive mutations, otherwise they are false positive mutations.

[0108] Table 3 Filtration thresholds of blood samples

[0109]

[0110] In the above table, HOT hotspots come from the public database OncoKB, and the database URL is: https: / / www.oncokb.org.

[0111] Validation example: A method for filtering sequencing background noise

[0112] The blood or tissue samples compiled by our company as true positive mutation data sets and false positive mutation data sets have been sequenced and manually reviewed to ensure accuracy.

[0113] The method of the embodiment is used to filter the background noise of the sequencing sample to verify the accuracy. The experimental process is as follows: Figure 2 As shown, the details are as follows:

[0114] (1) Before performing mutation filtering, prepare the background baseline file, the repeated region file, and the trained repeated region filtering model file.

[0115] (2) After removing the sequencing adapters and low-quality reads, the sample fastq files obtained by the second-generation sequencing are aligned to the reference genome to obtain the aligned BAM files. Using commonly used point mutation and insertion and deletion mutation identification tools such as vardict and varscan (commonly used tools include: GATK, samtools, Mutect, etc.), the location information of all identified point mutations and insertion and deletion mutations on the genome, the reference base sequence and the mutated base sequence, the mutant allele frequency (AF), the number of sequencing reads supporting the mutation (AD), and the number of all sequencing reads covering the mutation position (DP) are obtained.

[0116] (3) For the candidate mutation to be filtered, the BAM file compared to the reference genome is PCR deduplicated. Based on the deduplicated BAM, the number of all paired-end reads supporting the mutation and the number of all reads supporting the mutation are counted.

[0117] (4) For the candidate mutations to be filtered, annotate them according to the hotspot mutations to determine whether the mutation is a hotspot mutation.

[0118] (5) For the candidate mutations to be filtered, the background abundance of the candidate mutations is determined according to the established background baseline file, and the ratio of the actual mutation abundance (AF) to the average abundance of the mutation in the background baseline (AD.sum / DP.sum), i.e., ratio.sum, and the ratio of the actual mutation abundance (AF) to the maximum abundance of the mutation in the background baseline (AD.max / DP.max), i.e., ratio.max, are calculated. The Fisher one-sided test p-value of the four values ​​of the actual number of reads supporting the mutation (AD), the actual mutation depth (DP), the background depth sum (DP.sum), and the background support mutation sum (AD.sum), i.e., pvalue.sum.

[0119]

[0120] (6) For candidate mutations to be filtered, the repeat region file is used to determine whether they meet the initial screening conditions. If they meet the conditions, they are identified as false positive mutations. If they do not meet the conditions, the next step of identification is performed. The initial screening conditions are as follows: If the mutation meets: a. Insertion-deletion mutation; b. For deletion mutations, the reference base sequence of the mutation is required to have at least one base overlap with the repeat region of the intron region; For insertion mutations, the starting base position of the repeat region of the intron region is required to be ≤ the base position of the mutation on the chromosome ≤ the ending base position of the repeat region of the intron region, or the base position of the mutation on the chromosome is one position before the starting base position of the repeat region of the intron region; c. The bases of the insertion-deletion mutation are completely composed of tandem repeats of the minimum repeating unit of the repeat region, or the bases of the insertion-deletion mutation are part of the repeat unit.

[0121] (7) Using the repeat region file, determine whether the candidate mutation is located within 10bp of the repeat region and its upstream and downstream. If it is within 10bp of the repeat region and its upstream and downstream, calculate the characteristic value and use the repeat region filtering model to judge false positive and true positive, otherwise proceed to the next step of identification. For candidate mutations that meet the conditions, calculate the minimum number of bases of the nearest repeat region, the length of the insertion and deletion fragment, and the matching score of the insertion and deletion fragment and the repeat region. Combined with the ratio.sum value calculated in the above steps, the number of all reads supporting the mutation (after PCR deduplication), the number of all double-end reads supporting the mutation (after PCR deduplication), and the mutation abundance characteristic value, use the repeat region filtering model to identify false positive and true positive mutations.

[0122] (8) For candidate mutations, specific threshold filtering is performed using the ratio.sum, pvalue.sum, AF, and whether it is a hotspot mutation calculated in the above steps. Mutations that meet the filtering conditions are considered true positive mutations, and those that do not meet the filtering conditions are considered false positive mutations.

[0123] Among them, the above steps (3), (4), and (5) are for calculating the characteristic values ​​of the candidate mutations. There is no necessary order, and the order can be adjusted arbitrarily or calculated in parallel.

[0124] The above method was used to evaluate the performance of the false positive recognition model for the repeat region of the sequencing data of 11,944 blood samples. The recognition effect of the model on the true positive and false positive mutations in the repeat region is as follows: Figure 3 As shown in Table 4:

[0125] Table 4 Identification effect of blood samples

[0126] Evaluation Metrics Score Accuracy 0.999534 Recall 0.999499 F-Measure 0.999534 Matthews correlation coefficient (MCC) 0.999068

[0127] The above method was used to evaluate the performance of the false positive recognition model for the repeat region of the sequencing data of 15715 tissue samples. The recognition effect of the model on the true positive and false positive mutations in the repeat region is as follows: Figure 4 As shown in Table 5:

[0128] Table 5 Recognition effect of tissue samples

[0129]

[0130]

[0131] from Figure 3 and Figure 4 It can be seen from the confusion matrix diagram that the model of blood and tissue samples predicts true positive as true positive and false positive as false positive at a rate of more than 99.9%, and the rate of false positive as false positive and false positive as true positive is between 0.1% and 0.2%, indicating that the model is very accurate in predicting true positive and false positive mutations, and has a very low misjudgment rate for true positive and false positive.

[0132] As shown in Tables 4 and 5, for the accuracy, recall, and F-value of the machine learning model, it is hoped that the model is close to 1, that is, it is hoped that the true positive is predicted as true positive as accurately as possible, and the false positive is predicted as false positive as accurately as possible. For the two models constructed in the embodiment, the accuracy, recall, and F-value are all above 0.999, which are very close to 1, and are considered to be good classification and recognition models.

[0133] Given that the Matthews correlation coefficient describes the correlation coefficient between the actual classification and the predicted classification, when it takes a value of 1, it means that the model can make a perfect prediction, and when it takes a value of 0, it means that the prediction result is worse than the random prediction result. Accordingly, the values ​​of the blood and tissue sample models constructed in the embodiment are very close to 1, which also proves to be a better classification and recognition model.

[0134] Experimental Example 1: A method for filtering sequencing background noise

[0135] The difference from the embodiment is that in the background noise filtering step, only the first step is used to filter mutations and insertion / deletion mutations and repeat regions.

[0136] The false positive mutation identification results in 15715 tissue samples and 11944 blood samples are summarized in Table 6.

[0137] Table 6 False positive mutation identification effect

[0138] Statistics organize blood Filtering the number of false positive mutations 40001 42463 Total false positive mutations 1112894 3480447 Proportion 3.59％ 1.22％

[0139] The results of identifying true positive mutations in 15,715 tissue samples and 11,944 blood samples are summarized in Table 7.

[0140] Table 7 True positive mutation identification effect

[0141] Statistics organize blood Filter the number of true positive mutations 4681555 2676534 Total number of true positive mutations 1851 831 Proportion 0.04％ 0.03％

[0142] Experimental Example 2: A method for filtering sequencing background noise

[0143] The difference from the embodiment is that in the background noise filtering step, only the second step is used to filter mutations and insertion / deletion mutations and repeat regions.

[0144] The false positive mutation identification results in 15715 tissue samples and 11944 blood samples are summarized in Table 8.

[0145] Table 8 False positive mutation identification effect

[0146] Statistics organize blood Filtering the number of false positive mutations 656417 1284611 Total false positive mutations 1112894 3480447 Proportion 58.98％ 36.91％

[0147] The results of identifying true positive mutations in 15,715 tissue samples and 11,944 blood samples are summarized in Table 9.

[0148] Table 9 True positive mutation identification effect

[0149] Statistics organize blood Filter the number of true positive mutations 4681555 2676534 Total number of true positive mutations 629 192 Proportion 0.01％ 0.01％

[0150] Experimental Example 3: A method for filtering sequencing background noise

[0151] The difference from the embodiment is that in the background noise filtering step, only the third step is used to filter mutations and insertion / deletion mutations and repeat regions.

[0152] The false positive mutation identification results in 15715 tissue samples and 11944 blood samples are summarized in Table 10.

[0153] Table 10 False positive mutation identification effect

[0154] Statistics organize blood Filtering the number of false positive mutations 1048951 3426942 Total false positive mutations 1112894 3480447 Proportion 94.25％ 98.46％

[0155] The results of identifying true positive mutations in 15,715 tissue samples and 11,944 blood samples are summarized in Table 11.

[0156] Table 11 True positive mutation identification effect

[0157] Statistics organize blood Filter the number of true positive mutations 4681555 2676534 Total number of true positive mutations 231 696 Proportion 0.00％ 0.03％

[0158] Experimental Example 4: A method for filtering sequencing background noise

[0159] The difference from the embodiment is that in the background noise filtering step, only the first and second steps are used to filter mutations and insertion / deletion mutations and repeat regions.

[0160] The false positive mutation identification results in 15,715 tissue samples and 11,944 blood samples are summarized in Table 12.

[0161] Table 12 False positive recognition effect

[0162] Statistics organize blood Filtering the number of false positive mutations 680988 1297674 Total false positive mutations 1112894 3480447 Proportion 61.19％ 37.28％

[0163] The results of identifying true positive mutations in 15,715 tissue samples and 11,944 blood samples are summarized in Table 13.

[0164] Table 13 True positive mutation identification effect

[0165] Statistics organize blood Filter the number of true positive mutations 4681555 2676534 Total number of true positive mutations 2396 1016 Proportion 0.05％ 0.04％

[0166] Experimental Example 5: A method for filtering sequencing background noise

[0167] The difference from the embodiment is that in the background noise filtering step, only the second and third steps of filtering are used to filter mutations and insertion / deletion mutations and repeat regions.

[0168] The false positive mutation identification results in 15,715 tissue samples and 11,944 blood samples are summarized in Table 14.

[0169] Table 14 False positive mutation identification effect

[0170] Statistics organize blood Filtering the number of false positive mutations 1074343 3456321 Total false positive mutations 1112894 3480447 Proportion 96.54％ 99.31％

[0171] The identification results of true positive mutations in 15,715 tissue samples and 11,944 blood samples are summarized in Table 15.

[0172] Table 15 True positive mutation recognition effect

[0173] Statistics organize blood Filter the number of true positive mutations 1074343 3456321 Total number of true positive mutations 834 805 Proportion 0.02％ 0.03％

[0174] Experimental Example 6: A method for filtering sequencing background noise

[0175] A total of 15,715 tissue samples and 11,944 blood samples were collected, and the sequencing background noise was filtered using the method of the embodiment. The recognition effect of false positives in the point mutations and insertion / deletion mutations and repeat regions of the samples is shown in Table 16.

[0176] Table 16 False positive recognition effect

[0177] Statistics organize blood Filtering the number of false positive mutations 1098913 3469382 Total false positive mutations 1112894 3480447 Proportion 98.74％ 99.68％

[0178] A total of 15,715 tissue samples and 11,944 blood samples were collected, and the sequencing background noise was filtered using the method of the embodiment. The recognition effect of true positives in the point mutations and insertion / deletion mutation repeat regions of the samples is shown in Table 17.

[0179] Table 17 True Yang Identification Effect

[0180] Statistics organize blood Filter the number of true positive mutations 1074343 3456321 Total number of true positive mutations 2601 1629 Proportion 0.06％ 0.06％

[0181] It can be seen from the recognition effects of Experimental Examples 1 to 6 that for the repetitive regions of point mutations and insertion / deletion mutations, in the background noise filtering step of the method for filtering the sequencing background noise in the embodiment, only the first step is used for filtering, and only 3.59% of the false positive mutations in tissue samples and only 1.22% of the false positive mutations in blood samples can be identified and filtered, and the filtering effect is poor; only the second step is used for filtering, 58.98% of the false positives in tissue samples and 36.91% of the false positives in tissue samples can be identified and filtered, and the filtering effect is better than that of only using the first step for filtering; only the third step is used for filtering, 94.25% of the false positive mutations in tissue samples and 98.46% of the false positive mutations in blood samples are identified and filtered, with better filtering effect.

[0182] By comparing the filtering effects after using the first step + the second step, the second step + the third step, and the first step + the second step + the third step (i.e. the three steps added together), it was found that the combined filtering effect of the first step + the second step was better than the effect of running the two steps alone. The combined filtering effect of the second step + the third step also significantly optimized the performance of running the second step alone. The filtering effect after adding the three steps was the best, which could achieve a false positive filtering rate of 98.74% for tissue samples and 99.68% for blood samples.

[0183] Judging from the true positive filtering ratio at each step, the misjudgment ratio for true positive mutations at each step is relatively low. While the false positive filtering rate increases, there will be a small increase in the true positive misjudgment rate, but the overall true positive identification accuracy is also above 99%.

[0184] Experimental Example 7: A method for filtering sequencing background noise

[0185] Since in the method for filtering sequencing background noise in the embodiment, the second step of background noise filtering only filters mutations located in the repeat region and within 10bp upstream and downstream thereof, and the third step can filter all mutations, at the same time, false positive mutations are more likely to occur near the repeat region. Therefore, the second and third steps are used to filter mutations located in the repeat region and within 10bp upstream and downstream thereof, respectively, to compare the difference in filtering effects of the second and third steps for the above-mentioned key areas. The recognition effect is shown in Tables 18 and 19.

[0186] 15,715 tissue samples and 11,944 blood samples were collected, and point mutations and insertion / deletion mutations in the repetitive region and mutations within 10 bp upstream and downstream thereof were filtered using the second and third steps of the background noise filtering step of the method for filtering sequencing background noise in the embodiment. The false positive mutation recognition effect of the samples is shown in Table 18.

[0187] Table 18 Comparison of false positive mutation recognition effects

[0188] project Step 2: Filter false positive mutations Step 3: Filter false positive mutations organize 95.51％(656416 / 687251) 91.82％(631025 / 687251) blood 98.64％(1284611 / 1302263) 96.39％(1255232 / 1302263)

[0189] A total of 15,715 tissue samples and 11,944 blood samples were collected. The second and third steps of the background noise filtering step of the method for filtering sequencing background noise in the embodiment were used to filter point mutations and insertion / deletion mutations in the repetitive region and mutations within 10 bp upstream and downstream. The true positive mutation recognition effect of the samples is shown in Table 19.

[0190] Table 19 Comparison of True Yang Identification Effects

[0191] project The second step is to filter out true positive mutations Step 3: Filter true positive mutations organize 0.00％(29 / 581472) 0.00％(26 / 581472) blood 0.06％(192 / 325960) 0.03％(83 / 325960)

[0192] Only the false positive mutation filtering within the repeat region and the upstream and downstream 10bp range was tested, and it was found that the false positive mutation filtering effect of step two was better than that of step three, and the recognition rate of true positive was above 99%.

[0193] In summary, for point mutations and insertion-deletion mutations, the method for filtering sequencing background noise in the embodiment of the present invention showed excellent accuracy in identifying false positive mutations and true positive mutations.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for constructing a repeated region background noise filtering model, characterized in that: The following steps are involved: (1) performing next-generation sequencing on the sample and obtaining the SNV and indel mutations of the sample by comparing it with the reference genome; (2) extracting low-complexity regions and a simple repeat region list from the reference genome repeat region list; identifying simple repeat sequences on the reference genome; merging the simple repeat sequence region with the repeat region list according to base positions to obtain a repeat region list; and determining a training set according to the repeat region list; The training set is the list of repeat regions, each repeat region is extended 10 bp upstream and downstream, and true positive mutations and false positive mutations on exons with similar numbers are specified within this range; Includes methods for annotating repetitive regions including intronic regions: According to the specified transcript, find the Exon position corresponding to the transcript on GENECODE, and annotate the merged repeat region list; The repeat region of the intron region is a simple repeat region in which the start and end base positions of the repeat region are on the intron and are more than 2 bp away from the exon, and the minimum repeat unit is repeated 5 times or more in series; (3) constructing a random forest model for identifying true positive and false positive mutations using the training set according to the characteristic values, thereby completing the model construction; The characteristic values ​​include: the length of the mutation; AF; ratio.sum value; The number of all paired-end reads supporting the mutation; The number of all reads supporting the mutation; Wherein, AF is the actual mutation abundance of the mutation to be filtered; the ratio.sum value is the ratio of AF to the average abundance of the mutation in the background baseline; the average abundance is the ratio of AD.sum to DP.sum; the AD.sum value is the sum of the number of sequencing sequences that support the mutation in the sample; the DP.sum value is the sum of the number of sequencing sequences that cover the mutation; When the mutation is a snv mutation, the length of the mutation is equal to 0; The characteristic values ​​also include: The minimum distance between the mutated base position and the nearest repeat region; or The indel base matches the nearest repeated base in the repeated region, and the calculation formula of the matching score is: score / length.

2. The method for constructing a background noise filtering model according to claim 1, characterized in that: The simple repeat region refers to a repeat region with a repeat unit ≤ 6 bp.

3. The method for constructing the repeated region background noise filtering model according to claim 1, characterized in that: In step (1), the samples are negative samples, and the number of samples is not less than 40.

4. The method for constructing the repeated region background noise filtering model according to claim 1, characterized in that: In the step (1), after obtaining the snv and indel mutations of the sample, the DP.sum value, AD.sum value, DP.max value, and AD.max value of the sample are counted; Among them, the DP.max value is the maximum number of sequencing sequences in the sample that cover the mutation; the AD.max value is the maximum number of sequencing sequences in the sample that support the mutation.

5. The method for constructing the repeated region background noise filtering model according to claim 4, characterized in that: In the step (1), a filtered background baseline is constructed; the evaluation value of the background baseline includes at least one of the following: ratio.sum value; ratio.max value; the ratio.max value is the ratio of the maximum abundance of the mutation in AF to the background baseline; the maximum abundance is the ratio of AD.max to DP.max; pvalue.sum value; the pvalue.sum value is the Fisher one-sided test p-value value of AD, DP, DP.sum, and AD.sum; Among them, the AD is the number of sequencing sequences that actually support mutations of the mutation to be filtered; and the DP is the number of sequencing sequences that actually cover mutations of the mutation to be filtered.

6. The method for constructing the repeated region background noise filtering model according to claim 4, characterized in that: When a mutation does not exist in the background baseline, the DP.sum and DP.max are given a virtual value of 100000, and the AD.sum and AD.max are given a virtual value of 1 for calculation.

7. The method for constructing the repeated region background noise filtering model according to claim 1, characterized in that: The minimum distance between the mutated base position and the nearest repeat region ranges from 0 to 10. If the mutated base position is within the repeat region, the minimum distance is 0.

8. A device comprising the repetitive region background noise filtering model constructed by the method of claim 1.

9. A method for filtering sequencing background noise, characterized in that: include: Identify true positive and false positive mutations of the sample, and filter out false positive mutations; The filtering adopts characteristic values, and the characteristic values ​​include: The number of bases in the nearest repeat region; the length of the bases of the indel; the matching score between the bases of the indel and the repeat region; the ratio.sum value; the number of all reads supporting the mutation; the number of all paired-end reads supporting the mutation; AF; Wherein, AF is the actual mutation abundance of the mutation to be filtered; the ratio.sum value is the ratio of AF to the average abundance of the mutation in the background baseline; the average abundance is the ratio of AD.sum to DP.sum; the AD.sum value is the sum of the number of sequencing sequences that support the mutation in the sample; the DP.sum value is the sum of the number of sequencing sequences that cover the mutation; The characteristic values ​​also include: The minimum distance between the mutated base position and the nearest repeat region; or The indel base matches the nearest repeated base in the repeated region, and the calculation formula of the matching score is: score / length.

10. The method for filtering sequencing background noise according to claim 9, characterized in that: The filtering includes the following conditions: 1) For deletion mutations, the reference base sequence of the mutation is required to overlap with the intronic repeat region by at least one base; for insertion mutations, the starting base position of the intronic repeat region is required to be ≤ the base position of the mutation on the chromosome ≤ the ending base position of the intronic repeat region, and / or the base position of the mutation on the chromosome is the position before the starting base position of the intronic repeat region; and 2) The bases of the mutation indel are completely composed of tandem repeats of the minimum repeating unit of the repeat region, or the bases of the indel are part of the repeating unit.

11. A method for filtering sequencing background noise, characterized in that: include: Filter out false positive mutations based on thresholds; The characteristic values ​​of the threshold include: AF, ratio.sum value, pvalue.sum value; Among them, AF is the actual mutation abundance of the mutation to be filtered; the ratio.sum value is the ratio of AF to the average abundance of the mutation in the background baseline; the average abundance is the ratio of AD.sum to DP.sum; the AD.sum value is the sum of the number of sequencing sequences supporting the mutation in the sample; the DP.sum value is the sum of the number of sequencing sequences covering the mutation; the AD is the number of sequencing sequences that actually support the mutation to be filtered; the DP is the number of sequencing sequences that actually cover the mutation to be filtered; The pvalue.sum value is the Fisher one-sided test p-value of AD, DP, DP.sum, and AD.sum; The characteristic values ​​also include: The minimum distance between the mutated base position and the nearest repeat region; or The indel base matches the nearest repeated base in the repeated region, and the calculation formula of the matching score is: score / length.

12. The method for filtering sequencing background noise according to claim 11, characterized in that: The filtering also includes the following conditions: 1) For deletion mutations, the reference base sequence of the mutation is required to overlap with the intronic repeat region by at least one base; for insertion mutations, the starting base position of the intronic repeat region is required to be ≤ the base position of the mutation on the chromosome ≤ the ending base position of the intronic repeat region, and / or the base position of the mutation on the chromosome is the position before the starting base position of the intronic repeat region; and 2) The bases of the mutation indel are completely composed of tandem repeats of the minimum repeating unit of the repeat region, or the bases of the indel are part of the repeating unit.

13. The method for filtering sequencing background noise according to claim 11, characterized in that: The characteristic values ​​also include: The number of bases in the nearest repeat region; the length of the bases of the indel; the matching score between the bases of the indel and the repeat region; the number of all reads supporting the mutation; the number of all double-end reads supporting the mutation.

14. The method for filtering sequencing background noise according to claim 11, characterized in that: The calculation formula of the pvalue.sum value is as follows:

15. The method for filtering sequencing background noise according to claim 11, characterized in that: When the sample is a tissue sample and the mutation type is SNV or indel, the following conditions are used to filter out false positive mutations: for HOT sites, the threshold value AF≥1% is used for filtering; for non-HOT sites, the threshold value AF≥2% is used for filtering.

16. The method for filtering sequencing background noise according to claim 11, characterized in that: When the sample is a tissue sample, the mutation type is SNV or indel and there is a background baseline, the following conditions are used to filter to remove false positive mutations: for HOT sites, the threshold ratio.sum is filtered at ≥5; for non-HOT sites, the threshold ratio.sum is filtered at ≥10.

17. The method for filtering sequencing background noise according to claim 11, characterized in that: The sample is a blood sample. When the mutation type is SNV or indel, the sample is filtered according to the following conditions to remove false positive mutations, and the sample is filtered according to the threshold value pvalue.sum≤0.0001.

18. The method for filtering sequencing background noise according to claim 11, characterized in that: The sample is a blood sample. When the mutation type is SNV or indel, it is filtered according to the following conditions to remove false positive mutations. For HOT sites, it is filtered according to the threshold AF≥0.1%; for non-HOT sites, it is filtered according to the threshold AF≥0.2%.

19. The method for filtering sequencing background noise according to claim 11, characterized in that: The sample is a blood sample. When the mutation type is SNV or indel and there is a background baseline, it is filtered according to the following conditions to remove false positive mutations. For HOT sites, it is filtered according to the threshold ratio.sum≥5; for non-HOT sites, it is filtered according to the threshold ratio.sum≥10.

Images