Background noise mutation filtering method for detecting ctDNA through ultra-deep sequencing, computer medium and computer system

Abstract

The invention relates to the technical field of bioinformatics, in particular to a background noise mutation filtering method for detecting ctDNA through ultra-deep sequencing, a computer medium and a computer system. According to the method, on non-target areas in a section of area extending towards the left side and the right side of all target sites, according to mutation direction classification, the probability that the corresponding mutation direction is background mutation is counted as a baseline, and a model is constructed based on the target site sequencing depth and the background mutation probability; and calculating the maximum mutation times when the mutation is the background mutation as a threshold value for distinguishing the real mutation and the background mutation. The method is different from a traditional background noise filtering mode which depends on a normal sample to construct a baseline, the baseline can be constructed only by using the to-be-detected sample, the filtering model is constructed, and real mutation and background mutation are well distinguished; meanwhile, according to the method, performance indexes of different sensitivity and specificity can be realized through specific parameter adjustment, so that the requirements of different application scenes are met.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and in particular to a background noise mutation filtering method, a computer medium, and a computer system for ultra-deep sequencing detection of ctDNA. Background Art

[0002] The detection of circulating tumor DNA (ctDNA) as a tumor marker has shown significant advantages in modern medicine. It avoids the trauma and risks caused by traditional tissue biopsy, provides great convenience for patients, and allows doctors to conduct frequent and safe disease monitoring and keep abreast of the development of tumors. It can provide valuable information for early diagnosis, evaluation of treatment effects, and monitoring of disease recurrence.

[0003] Single nucleotide variants (SNVs) and insertion / deletion mutations (InDels) are common types of variation in the human genome. Many SNV and InDel mutations can cause abnormal and uncontrolled cell growth, affecting the steady-state development of a series of key cellular functions. Accurate detection and identification of SNVs and InDels play a very important role in clinical diagnosis, medication guidance, treatment and prognosis. With the continuous development of sequencing technology and bioinformatics, more and more sequencing platforms and mutation identification algorithms are used to detect SNVs and InDels.

[0004] The content of ctDNA in blood circulation is extremely low. The detection of single nucleotide mutations and insertion and deletion mutations in ctDNA not only requires extremely high sample enrichment and purification technology, but also the incorrect binding of PCR polymerase during library amplification and enrichment and the sequencing errors of the sequencer itself, which introduces high background noise problems in ultra-deep sequencing technology. Distinguishing between real mutations and noise is also a major problem. For example, due to the technical limitations of the commonly used second-generation sequencing platforms, false positive results often appear in the final analysis results, including false positive noise caused by a large number of sequencing platforms. In order to achieve accurate detection of low-frequency mutations in plasma ctDNA, the current relevant technologies include: single molecule label (Unique Molecular Identifier, UMI) technology, or the use of some false positive removal tools (such as GATK VariantFiltration, fpfilter), etc. Using UID (uniqueidentifier) ​​tags for forward and reverse double-strand error correction can correct sequencing errors, but it cannot distinguish between real mutations and background noise mutations. Traditional false positive removal tools usually require a certain number of normal plasma samples as a baseline. In particular, different detection panels have different algorithms for controlling background noise. Fixed panels can use a large number of normal samples to build a baseline at one time, or use machine learning algorithms to distinguish between "real" mutations and background noise. When a personalized panel detects a single sample, it can also use a dozen normal samples to build a baseline to obtain higher sensitivity and specificity, but the cost will also increase significantly, increasing the difficulty of practical application. Summary of the invention

[0005] In order to overcome the defects of the prior art, one of the objectives of the present invention is to provide a background noise mutation filtering method for detecting ctDNA by ultra-deep sequencing, which does not require normal samples to construct a baseline, and can distinguish the mutations detected at the target site as "real mutations" and background noise mutations under ultra-deep sequencing.

[0006] A second object of the present invention is to provide a computer medium for executing the background noise mutation filtering method for ultra-deep sequencing detection of ctDNA provided by the present invention.

[0007] The third object of the present invention is to provide a computer system for running the background noise mutation filtering method for ultra-deep sequencing detection of ctDNA provided by the present invention.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A method for filtering background noise mutations in ultra-deep sequencing of ctDNA, comprising:

[0010] 1) Obtaining sequencing data of the sample to be tested;

[0011] 2) Perform sequence quality control and reference genome comparison;

[0012] 3) Obtain the base mutation results at each position;

[0013] 4) Constructing a baseline: Classify and count by mutation direction, and count the mutation frequency distribution characteristics of each site in a region on both sides of the target site, excluding the target site;

[0014] 5) Constructing a background noise mutation filtering model: According to the mutation direction, a background noise mutation filtering model is constructed based on the sequencing depth n of the target site and the mutation frequency distribution characteristics of the mutation direction, and the maximum number of mutations caused by background noise in the mutation direction is determined as the filtering condition.

[0015] In a specific embodiment of the present invention, in step 4), the background noise mutation probability p of different mutation directions in the 10-100 bp region on both sides of the target site is counted. m , where m represents the mutation direction.

[0016] As a preferred method, in step 4), the background noise mutation probability p of different mutation directions in the 40-60 bp region on both sides of the target site is counted. m .

[0017] In a specific embodiment of the present invention, the background noise mutation probability p in different mutation directions is m is the quantile of all mutation abundances in the mutation direction.

[0018] In a specific embodiment of the present invention, the specific method of step 5) constructing a background noise mutation filtering model is: constructing an inverse cumulative distribution function model, and statistically analyzing the probability P of the site mutation being a background noise mutation: P(X≥k)≤α, where X~B(n,p m ), X obeys n and p m The determined k value represents the maximum number of mutations caused by background noise.

[0019] The number of mutations in the target site in this mutation direction is counted. If the number of mutations is greater than k, it is considered a true mutation, otherwise it is background noise.

[0020] Preferably, α=0.01; Quantile=80; and the target site sequencing depth n≥10,000X.

[0021] Optionally, the sample to be tested is a body fluid sample or a tissue sample; further preferably, the sample to be tested is a plasma sample;

[0022] Optionally, the sequencing data of the sample to be tested is the sequencing data of the sample to be tested after targeted library construction, wherein the targeted library construction includes constructing a library containing the target region by designing specific amplification primers through multiple amplification; and also includes obtaining the library containing the target region by probe capture.

[0023] A computer medium comprising:

[0024] 1) Sequence quality control module: used to remove low-quality sequences, remove duplicates and correct sequencing errors in sequencing data;

[0025] 2) Sequence alignment module: used to align the sequencing data output by the sequence quality control module with the human reference genome;

[0026] 3) Mutation result identification and mutation feature statistics module: used to identify the sequencing depth of each site, the number of mutations and mutation frequencies corresponding to the mutation of the base at that position to the other three non-reference bases;

[0027] 4) Background noise judgment and filtering module: used to execute the background noise filtering model and calculate the maximum number of mutations k in the mutation direction caused by background noise; when the number of mutations in the mutation direction of the target site is greater than k, it is judged as a real mutation, otherwise it is background noise;

[0028] The background noise filtering model is constructed in a region on both sides of the target site according to the mutation direction, based on the sequencing depth n of each site and the mutation frequency distribution characteristics of the mutation direction, to determine the maximum number of mutations caused by background noise in the mutation direction.

[0029] Optionally, the background noise filtering model implemented by the background noise judgment and filtering module is an inverse cumulative distribution function model P(X≥k)≤α, where P is the probability of a mutation being a background noise mutation; X~B(n,p m ), X obeys n and p m Binomial distribution, n is the sequencing depth of the target site, p m is the background noise mutation probability in different mutation directions, where m represents the mutation direction, and when the significance level = α, the determined k value represents the maximum number of mutations caused by background noise; where p m It is the statistical data of each mutation site within the 10-100 bp region (more preferably, within the 40-60 bp region) on both sides of the target mutation site, excluding the target site;

[0030] A computer system comprises a processor and a memory; the processor and the memory are communicatively connected, wherein the memory is used to store a computer program, the processor is used to call the computer program, the computer program comprises program instructions, and when the program instructions are executed by the processor, the above method is executed.

[0031] Beneficial effects of the present invention:

[0032] 1) The present invention classifies the mutation results detected in the non-target area extending to the left and right sides of all target sites according to the mutation direction, and statistically calculates the probability of the corresponding mutation direction being a background mutation as a baseline. A model is constructed based on the sequencing depth of the target site and the probability of background mutation, and the maximum number of mutations that are background mutations is calculated as the threshold for distinguishing true mutations from background mutations. Different from the traditional background noise filtering method that relies on normal samples to build a baseline, only the sample to be tested itself can be used to build a baseline and a filtering model, which can well distinguish true mutations from background mutations;

[0033] 2) Furthermore, the method of the present invention can achieve performance indicators of different sensitivity and specificity through specific parameter adjustment to meet the needs of different application scenarios. DETAILED DESCRIPTION

[0034] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0035] Unless otherwise specified, the techniques used and covered herein are standard methods known to those skilled in the art to which the present invention belongs. The materials, methods and examples are only used for illustrative purposes and are not intended to limit the scope of the present invention in any way.

[0036] As described herein, the original sequencing data is obtained by preparing a sequencing library and performing upper-level sequencing on the nucleic acid sequence of the sample to be tested; including obtaining the nucleic acid of the sample to be tested, preparing a sequencing library of the nucleic acid, and the preparation method of the sequencing library is performed according to the requirements of the selected sequencing method. The sequencing library can be prepared by an amplicon library construction method, wherein the amplicon library construction includes constructing a library containing a target region by multiple amplification by designing specific amplification primers; and also includes obtaining a library containing a target region by a probe capture method; the HiSeq2000 / 2500 sequencing platform of Illumina can be selected but not limited to, and the sequencing method can be selected by single-end sequencing or double-end sequencing, and the obtained offline data are fragments read out, called reads;

[0037] The present invention relates to computer media or computer system products, specifically to computer-readable storage media on which are loaded computer-readable program instructions for executing the present invention; wherein "computer-readable storage media" refers to a tangible device that can maintain and store instructions used by an instruction execution device.

[0038] Example 1

[0039] This embodiment provides a method for filtering background noise mutations in ctDNA by ultra-deep sequencing, and the specific steps are as follows:

[0040] (1) Sequencing data comparison and result statistics: After obtaining the sequencing data of the sample to be tested for sequence quality control and comparison with the reference genome, the base mutation results at each position are obtained;

[0041] (2) Build a baseline:

[0042] S1: All target sites extend to the left and right sides as non-target areas, and all mutation detection results in these areas are counted;

[0043] S2: Classify according to mutation direction, and use the quantile Q of mutation abundance in each mutation direction as the background noise mutation probability in that direction;

[0044] S3: A baseline file is constructed for each sample, including all mutation directions and the corresponding background noise mutation probability p m , where m represents the mutation direction.

[0045] (3) Constructing the filtering model

[0046] Call the inverse cumulative distribution function (ICDF) to build a filtering model P(X≥k)≤α, where X~B(n,p m ), X obeys n and p m Binomial distribution, n is the sequencing depth, p m is the background noise mutation probability in different mutation directions;

[0047] (4) Identifying true mutations

[0048] For each target mutation site, a filtering model is used to calculate the maximum number of mutations k expected to be caused by background noise when the sequencing depth is n at the significance level α; when the number of mutations is greater than k, the mutation at the site is considered to be a "real" mutation.

[0049] Example 2 Design of experiment to verify the feasibility of the method of the present invention

[0050] 1. Experimental design

[0051] S1. Sample preparation plan

[0052] Three cell lines were used for dilution in three gradients as positive samples, and WBC and normal human plasma were combined as negative samples;

[0053] Table 1. Experimental sample information

[0054]

[0055] S2. Screening of target sites

[0056] Based on the mutation results detected for each cell line in public databases, hot spots, driver mutations, tumor-specific mutations, and sites with high mutation abundance were preferentially selected as target mutations.

[0057] Table 2. Number of sites detected in cell lines

[0058]

[0059] 2. Test process:

[0060] S1. Experimental Procedure

[0061] Primers were designed and synthesized according to the target mutation sites of each cell line, and the gradient dilution samples and negative samples were amplified using the primers of each cell line to construct sequencing libraries, which were then sequenced.

[0062] S2. Analysis Process

[0063] After sequencing sequence quality control, use bwa to align to the genome, the sequencing depth must reach 100,000X, correct sequencing errors, and then use bcftools to detect mutations.

[0064] S3. Statistics of effective sites detected

[0065] The target mutation site with a sequencing depth of ≥10,000X after sequence correction was defined as a valid site, and the effective positive sites in each mutation abundance interval of the gradient dilution sample, the corresponding negative sites in the negative sample, and the actually detected "negative" sites were statistically analyzed.

[0066] Table 3. Detection of effective sites in each mutation abundance interval

[0067]

[0068] S3: Building a baseline

[0069] All mutations extending L bp to the left and right sides of the target site are intercepted and classified according to the mutation direction. The quantile Q of the mutation abundance in each mutation direction is used as the background noise mutation probability p in that direction. m , where a baseline is constructed for each sample in all mutation directions.

[0070] S4: Filtering background noise

[0071] For each target site, a significance level α is given and the number of detections n is the sequencing depth. In Python, use binom.ppf(α,n,p m) function calculates the maximum number of mutations k expected to be caused by background noise under confidence = 1-α. When the number of mutations detected at the target site exceeds k, the site is considered to be a true mutation, otherwise it is background noise;

[0072] S5: Different extension lengths L on both sides of the target site, different quantiles Q used for the background noise mutation probability in each mutation direction, and different significance levels α of the binomial distribution can achieve filtering performance with different specificity and sensitivity. The data in the table below shows the filtering performance after using different extension lengths L to construct the baseline.

[0073] Table 4. Filtration performance

[0074]

[0075]

[0076] The results shown in Table 4 show that the sensitivity and specificity data of the cell lines and normal samples after filtering show that the present invention can achieve the specificity and sensitivity of the traditional normal sample baseline filtering background noise. It has the feasibility of distinguishing the real mutation and background noise mutation of the target site under ultra-deep sequencing, and can well control the false positive results while ensuring sufficient sensitivity.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for filtering background noise mutations in ultra-deep sequencing of ctDNA, characterized in that: include: 1) Obtaining sequencing data of the sample to be tested; 2) Perform sequence quality control and reference genome comparison; 3) Obtain the base mutation results at each position; 4) Constructing a baseline: Classify and count the mutation frequencies detected in a region on both sides of the target site, excluding the target site, by mutation direction; 5) Constructing a background noise mutation filtering model: According to the mutation direction, a background noise mutation filtering model is constructed based on the sequencing depth n of the target site and the mutation frequency distribution characteristics of the mutation direction, and the maximum number of mutations caused by background noise in the mutation direction is determined as the filtering condition.

2. The background noise mutation filtering method for detecting ctDNA by ultra-deep sequencing according to claim 1, characterized in that: In step 4), the background noise mutation probability p of different mutation directions in the 10-100 bp region on both sides of the target site is counted. m , where m represents the mutation direction.

3. The background noise mutation filtering method for detecting ctDNA by ultra-deep sequencing according to claim 2, characterized in that: In step 4), the background noise mutation probability p of different mutation directions in the 40-60 bp region on both sides of the target site is counted. m .

4. The method for filtering background noise mutations by ultra-deep sequencing of ctDNA according to any one of claims 1 to 3, characterized in that: The background noise mutation probability p of different mutation directions m is the quantile of all mutation abundances in the mutation direction.

5. The background noise mutation filtering method for detecting ctDNA by ultra-deep sequencing according to claim 4, characterized in that: Step 5) The specific method of constructing the background noise mutation filtering model is: construct an inverse cumulative distribution function model, and calculate the probability P of the site mutation being a background noise mutation: P(X≥k)≤α, where X~B(n,p m ), X obeys n and p m The binomial distribution of , significance level = α, the determined k value represents the maximum number of mutations caused by background noise; The number of mutations in the target site in this mutation direction is counted. If the number of mutations is greater than k, it is considered a true mutation, otherwise it is background noise.

6. The background noise mutation filtering method for detecting ctDNA by ultra-deep sequencing according to claim 5, characterized in that: α=0.01; Quantile=80; Target site sequencing depth n≥10,000×.

7. The method for filtering background noise mutations by ultra-deep sequencing for detecting ctDNA according to any one of claims 1 to 3 or 5 to 6, characterized in that: The sample to be tested is a body fluid sample or a tissue sample; Optionally, the sample to be tested is a plasma sample; Optionally, the sequencing data of the sample to be tested is the sequencing data of the sample to be tested after targeted library construction.

8. A computer medium, characterized in that include: 1) Sequence quality control module: used to remove low-quality sequences, remove duplicates and correct sequencing errors in sequencing data; 2) Sequence alignment module: used to align the sequencing data output by the sequence quality control module with the human reference genome; 3) Mutation result identification and mutation feature statistics module: used to identify the number of mutations and mutation frequencies corresponding to the mutation of each site base to the non-reference base; 4) Background noise judgment and filtering module: used to execute the background noise filtering model and calculate the maximum number of mutations k in the mutation direction caused by background noise; when the number of mutations in the mutation direction of the target site is greater than k, it is judged as a real mutation, otherwise it is background noise; The background noise filtering model is constructed in a region on both sides of the target site according to the mutation direction, based on the sequencing depth n of the target site and the mutation frequency distribution characteristics of the mutation direction, to determine the maximum number of mutations caused by background noise in the mutation direction.

9. The computer medium of claim 1, wherein: The background noise filtering model implemented by the background noise judgment and filtering module is an inverse cumulative distribution function model P(X≥k)≤α, where P is the probability of a mutation to a background noise mutation; X~B(n,p m ), X obeys n and p m Binomial distribution, n is the sequencing depth of the target site, p m is the background noise mutation probability in different mutation directions, where m represents the mutation direction, and when the significance level = α, the determined k value represents the maximum number of mutations caused by background noise; where p m It is the statistical data of each mutation site in the 10-100bp region on both sides of the target mutation site, excluding the target site; Optional, p m The statistical data of each site in the 40-60bp region on both sides of the target mutation site does not include the target site; Optional, p m is the quantile of all mutation abundances in the mutation direction; Optional, α=0.01; Quantile=80; The target site sequencing depth n≥10,000X.

10. A computer system, characterized in that: It comprises a processor and a memory; the processor and the memory are communicatively connected, wherein the memory is used to store a computer program, the processor is used to call the computer program, the computer program comprises program instructions, and when the program instructions are executed by the processor, the method according to any one of claims 1 to 7 is executed.

Images