Single-gene horizontal copy number variation recognition method, system and equipment based on whole exome sequencing queue and medium

Abstract

The invention discloses a single gene level copy number variation recognition method, system and equipment based on a whole exome sequencing queue and a medium. The method comprises the following steps: recognizing CNV at a single gene level by using WES short read length data; the method not only can identify the copy number variation of a single exon, but also can identify the copy number variation of a cross-exon, so that CNV identification at a single gene level is really realized, the problem of false positive caused by the difference of sequencing data is avoided, and the diagnosis accuracy is effectively improved. And diagnosis of rare single-gene genetic diseases is facilitated.

Description

Technical Field

[0001] The present invention belongs to the field of gene analysis technology and relates to a method for identifying gene copy number variations, and in particular to a method, system, device and medium for identifying single-gene copy number variations based on a whole-exome sequencing cohort. Background Art

[0002] Research indicates that there are 350 million people living with rare diseases worldwide, but only a small number can be accurately diagnosed and treated. Therefore, research into the diagnosis and treatment of rare, undiagnosed diseases holds significant clinical significance. Statistics show that approximately 80% of rare diseases, including various rare diseases, are caused by genetic factors and, in many cases, follow Mendelian inheritance patterns, thus being termed rare genetic diseases. Genomic genetic variation is the primary genetic factor contributing to the development of rare genetic diseases. Common genomic variations include simple nucleotide variants (SNVs), copy number variants (CNVs), and structural variants (SVs).

[0003] High-throughput whole-exome sequencing (WES) for SNV identification has been widely used for genetic disease screening and diagnosis, and has improved the diagnosis rate of rare genetic diseases in clinical practice. However, CNV analysis based on WES data has not been implemented in clinical scenarios. There are two main reasons. First, the sequencing range of WES data is discontinuous across the genome, and sequencing depth and coverage of different exon segments vary greatly due to factors such as sequence complexity and GC content. Second, differences in sequencing technology used, sequencing batch effects, and differences in sequencing samples can also lead to background differences in exon sequencing between different samples. These systematic differences make it very difficult to identify CNVs based on WES data. Therefore, it is urgent to improve analysis methods and increase accuracy to promote the wider use of WES data in clinical practice and further improve the diagnosis rate of genetic diseases.

[0004] Although using WES data to identify CNVs at a high-throughput level is not the best approach, CNV identification based on WES data can make full and effective use of the data. Currently, many tools have been developed to identify CNVs based on WES data, but almost all tools analyze CNVs by using different algorithms to evaluate coverage and sequencing depth for single sample data. However, due to the two reasons mentioned above, the data of a single sample itself has sequencing data uncertainty, which will result in relatively low sensitivity for existing tools to identify CNVs, and may result in a large number of false positives. In addition, the tools are limited by the parameters set by their models themselves, and the CNVs they identify are limited to specific targeted regions or to the background of the test data used in the algorithm testing process.

[0005] Therefore, the present invention utilizes a representative sequencing cohort using WES data from a number of different sequencing platforms, samples, and batches (sequencing contexts). This cohort is then used to train model parameters for CNV identification. The parameter-optimized CNV model is then used to identify CNVs in case WES data. This strategy allows CNV identification using optimal parameters based on the case's WES sequencing context, significantly reducing the number of false-positive results reported. Summary of the Invention

[0006] To expand the application of WES sequencing data and further improve the diagnosis rate of rare genetic diseases, the present invention provides a method, system, device, and medium for identifying single-gene copy number variations based on whole-exome sequencing cohorts. This method constructs a CNV identification model using WES sequencing data and uses the model to identify CNVs in WES data from individual cases. Furthermore, this method can be integrated into a computer system and combined with WES data storage media to form a specialized computing device for CNV identification.

[0007] The present invention adopts the following technical solutions:

[0008] A method for identifying single-gene copy number variations based on whole-exome sequencing cohorts, comprising the following:

[0009] Step 1: Collect WES sequencing data of individual samples and perform data quality control on the sample sequencing data, including base quality distribution, sequence length distribution, calculation of repeat sequence ratio, adapter sequence contamination, sequence GC content assessment, etc., to screen out qualified WES data for model construction.

[0010] Step 2: Preprocess qualified WES sequencing data (usually in FASTQ format), including sequence adapter removal, alignment to the human reference genome, marking duplicate sequencing reads, and base score correction. After preprocessing, each WES sequencing data file will be obtained after alignment (usually in BAM format).

[0011] Step 3: Read the comparison files of all qualified samples obtained in step 2 and input them into the unsupervised learning model. Through iterative optimization, the parameters are continuously adjusted to make the model better fit the data distribution, obtain the optimal baseline model and related model parameters for CNV identification, and obtain the best model.

[0012] Step 4: Apply the same quality control as in Step 1 to the WES sequencing data of the individual sample to be tested for CNVs. Then, perform sequence alignment in Step 2. Based on the aligned files and the optimal model generated in Step 3, perform CNV calling. Finally, report the CNV calling results as a text file or VCF (Variant Call Format) file.

[0013] In the above technical solution, the FASTQC tool can be used to analyze the quality control information of each sample in step 1. It should be noted that the WES sequencing data of the individual sample in step 1 is determined according to the model requirements. If a model with stronger universality is required, WES sequencing data of different sequencing backgrounds and more samples are used; if the model focuses more on the CNV identification accuracy, the WES data of a single sequencing background is used to build a model for CNV identification of WES sequencing data with the same sequencing background.

[0014] Furthermore, the preprocessing in step 2 may specifically include the following:

[0015] Step 2.1: Use Trimmomatic to remove adapters and filter low-quality bases to obtain a trimmed sequence file (usually in FASTQ format). This tool supports both single-end and paired-end sequencing data and can trim sequences based on a quality sliding window.

[0016] Step 2.2: Use the BWA tool to align the sequence file obtained in step 2.1 to the human reference genome to obtain an intermediate file (usually in BAM format). You can choose the common human reference genome version hg19 or hg38 as needed, and use the same version in subsequent analysis steps.

[0017] Step 2.3: Use the GATK (https: / / gatk.broadinstitute.org) tool to perform duplicate sequencing read length marking and base scoring correction steps on the intermediate files after the above alignment to obtain the final aligned files of each WES sequencing data (usually in BAM format).

[0018] Furthermore, step 3 is a key step of the technical solution of the present invention, which includes the following:

[0019] Step 3.1: Create an interval list. Create a list of intervals for CNV identification based on the exon coordinate annotation files on the human reference genome provided by different institutions, or based on the exon coordinate position annotation file (BED format) of the human reference genome (using the same human reference genome version used in Step 2.1).

[0020] Step 3.2: Preprocess the interval list. Considering the characteristics of WES sequencing data, its coverage area is usually within a certain range upstream and downstream of the exon, so padding space needs to be added to the created interval list. Exon intervals vary in size, so the interval list needs to be segmented, for example, into interval lists with large windows such as 500bp.

[0021] Step 3.3: Annotate the pre-processed and post-processed interval list. Based on the characteristics of the genomic sequence, annotate information such as the GC content of the interval sequence (i.e., the percentage of guanine G and cytosine C in the DNA sequence) and the sequence mappability (the degree to which the sequence can be uniquely mapped to the reference genome, reflecting the degree of repetitiveness and complexity of the regional sequence). This information will improve the accuracy of interval CNV identification.

[0022] Step 3.4: Using the preprocessed interval list, read the WES sequencing data alignment files of all samples used for training (i.e., obtained in step 2) and calculate the sequence read length (reads) coverage information for each interval in each sample interval list (the depth of sequencing reads covering the region within the specified genomic interval, i.e., the total number of aligned bases divided by the length of the target region).

[0023] Step 3.5: Based on the coverage information of all samples generated in step 3.4, the annotation information of the interval list in step 3.3 is used as a covariate for coverage correction. The interval list is further filtered based on the coverage distribution characteristics of the corrected samples, and abnormal coverage intervals outside the confidence interval are filtered out to form a new interval list; the removed intervals may be caused by repeated sequences, sequencing bias, or uneven enrichment efficiency. The basic steps of the filtering are: calculating the coverage of all exon target regions, statistically analyzing the distribution characteristics of the coverage, calculating the mean and standard deviation, filtering out regions outside the 95% confidence interval, and forming a new interval list for subsequent model construction.

[0024] Step 3.6: For the new interval list, calculate the ploidy of the sample within the interval to construct a baseline model for interval ploidy. The ploidy of the sample directly affects the coverage of different regions in the genome. The specific calculation steps are: using the sequencing coverage information of each sample in the exon interval in the new interval list, using principal component analysis (PCA) to correct the coverage of different samples to ensure that the ploidy estimation is not affected by technical bias and batch effects, using a Gaussian mixture model (GMM) to fit the coverage data, and determining the most likely ploidy based on maximum likelihood estimation (MLE) to obtain the baseline model and its parameters.

[0025] Step 3.7: Build a CNV prediction model. Using a cohort model (in which all training sample information is input simultaneously), combine the sample coverage information from step 3.4 and the ploidy baseline model information generated in step 3.6 as input features. Use unsupervised learning to train the model and obtain the optimal CNV recognition model parameters.

[0026] Furthermore, step 4 uses the model constructed in step 3 (as a general CNV identification model) to identify the CNV of the sample, and uses the CASE mode (referring to inputting the coverage information of a single sample that needs to identify CNV in the prediction stage) to perform CNV detection.

[0027] The present invention also provides a single gene level copy number variation identification system based on whole exome sequencing cohort, which is used to implement the method described in any of the above items.

[0028] The present invention further provides an electronic device, comprising:

[0029] one or more processors;

[0030] a memory for storing one or more programs;

[0031] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the above methods.

[0032] A computer-readable storage medium stores computer-executable instructions, wherein the instructions are used to implement any of the above methods when executed.

[0033] The beneficial effects of the present invention are: 1. The present invention proposes a new method that uses WES short-read data to identify CNVs at the single-gene level, which can be used for the clinical diagnosis of rare genetic diseases in clinical practice. 2. The method of the present invention can be applied to WES sequencing data of different depths or WES sequencing data from different platforms, and can also be extended to targeted sequencing and customized panel targeted sequencing. 3. This method overcomes the high false positive characteristic of using single sample data to identify CNVs due to the potential defects of WES sequencing data. The WES sequencing data queue is used to learn the common characteristics of WES sequencing data and form a model, avoiding reporting false positive results due to differences in the sequencing data itself, thereby improving the diagnostic accuracy. 4. This method can not only identify copy number variations of single exons but also identify copy number variations across exons, truly realizing CNV identification at the single-gene level, which is helpful for the diagnosis of rare single-gene genetic diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A complete flowchart of the copy number variation method at the single-gene level based on the whole-exome sequencing cohort.

[0035] Figure 2 .In the embodiment, a case visualization display of a model trained based on the 1000 Genomes Project WES sequencing data to identify single gene short fragment CNV (single exon copy number loss).

[0036] Figure 3.In the embodiment, a case visualization display of a model trained based on the 1000 Genomes Project WES sequencing data to identify single-gene long-fragment CNV (3 exon copy number loss). DETAILED DESCRIPTION

[0037] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] The present invention provides a method for identifying single-gene copy number variation based on a whole-exome sequencing cohort, and the technical solution includes:

[0039] Step 1: Collect WES sequencing data of individual samples, perform quality control on the sample sequencing data, and screen out qualified WES sequencing data;

[0040] Step 2: Preprocess the qualified WES sequencing data to obtain a file after each WES sequencing data is aligned with the human reference genome;

[0041] Step 3: All the alignment files obtained in step 2 are input into the unsupervised learning model. The parameters are continuously adjusted through iterative optimization to make the model better fit the data distribution, obtain the optimal baseline model and related model parameters for CNV identification, and obtain the best model;

[0042] Step 4: Perform the same data quality control as in step 1 on the WES sequencing data of the individual samples for CNV testing, then perform sequence alignment in step 2. Perform CNV identification based on the aligned files and the best model generated in step 3.

[0043] According to a specific embodiment of the present invention, Figure 1 The workflow of a corresponding computational model system provided by the present invention is presented, including steps such as data acquisition, tool preparation, model construction, prediction and inference, and identification result display. In order to further help understand the method of the present invention, the 1000genome project 3 ( ftp.1000genomes.ebi.ac.uk ) WES sequencing data is used as an example to illustrate the CNV model construction and related data processing steps in detail. According to the order of the technical solution implementation steps, the specific implementation steps of the embodiment are as follows:

[0044] Step S1: From the FTP server of the 1000 Genomes Project: ftp.1000genomes.ebi.ac.ukDownload the raw WES sequencing data (FASTQ format). To ensure the CNV model's recognition performance, preliminary evaluations suggest that WES sequencing data (FASTQ format) from at least 30 different individuals are required for the subsequent model building steps. Within the computational model system, the number of samples used for model training can be adjusted appropriately based on the WES data source and the computing power resources of the computing device. WES data from different sources can exhibit more pronounced technical bias and batch effects, so increasing the number of training samples is necessary to eliminate prediction bias caused by these complex sequencing background factors.

[0045] Step S2: Refer to the description in step 1 of the above technical solution, use the FASTQC tool to perform quality control on the downloaded WES sequencing data and screen out qualified samples for the next step. This step includes base quality distribution, sequence length distribution, calculation of repeat sequence ratio, adapter sequence contamination, sequence GC content assessment, etc. Specific quality control standards can be found according to the GATK official ( https: / / gatk.broadinstitute.org )’s suggestions: https: / / gatk.broadinstitute.org / hc / en-us / articles / 360035535912-Data-pre-processing-for-variant-discovery execution.

[0046] Step S3: Referring to step 2 of the technical solution, use Trimmomatic to remove the connectors and filter low-quality bases from the sequencing data of each sample (input FASTQ file) to obtain the trimmed sequence file (output FASTQ format).

[0047] Step S4: Refer to step 2 of the technical solution, use the alignment tool BWA to align the trimmed sequence data of all samples (input FASTQ file) to the human reference genome (using hg19 version), refer to https: / / github.com / lh3 / bwa , set the bwa tool parameters and execute.

[0048] Step S5: Refer to step 2 of the technical solution and input the intermediate BAM file obtained after alignment into the GATK tool to perform the MarkDuplicates and BaseRecalibrator steps. The specific execution parameters are based on the GATK recommendations. https: / / gatk.broadinstitute.org / hc / en-us / articles / 360035531652-FASTA- Reference-genome-format.

[0049] Step S6: Steps S1-S5 above will generate a qualified WES sequencing data alignment file for CNV model training, which will be used in subsequent model training. Create an interval list by referring to step 3 of the technical solution above. To do this, go to the UCSC Genome Browser (https: / / genome.ucsc.edu / cgi-bin / hgTables), select "Interval file define subsets of exonic regions," create the interval list, and download and save it to the file UCSC_RefGene_hg19.bed.

[0050] Step S7: Referring to step 3.2 of the technical solution, perform prediction processing on the interval list obtained in the above step, set the window parameter bin-length to 0, and the padding parameter to 250bp. The above parameter settings will cause the interval to be cut into 500bp interval lists; this step can be performed using the PreprocessIntervals function of gatk.

[0051] Step S8: Referring to step 3.3 of the technical solution, calculate the GC base content of the sequence (GC% = (G+C) / (A+T+C+G)*100%); use the BLAST tool to align all possible k-mers (k = 100 bp) with the human reference genome hg19, calculate their unique mapping rate (mappability), and annotate the above interval list. The step of annotating the intervals using the generated information can be implemented using the gatk function AnnotateIntervals.

[0052] Step S9: Referring to step 3.4 of the technical solution, based on the BAM file output in step S5, calculate the number of reads for all genomic intervals in the interval list file generated in step S7 to generate coverage information. Repeat this step to calculate coverage information for all samples and save this information to a readable text file (output in .tsv format) for each sample. This step can be implemented using the CollectReadCounts function of gatk.

[0053] Step S10: Referring to step 3.5 of the technical solution, read the coverage information for all samples generated in step S9 and perform linear regression to correct the coverage using the annotation information from step S8 as a coverage covariate. Then, based on the distribution characteristics of the corrected sample coverage, filter out intervals with coverage aberrations outside the confidence interval. Combine the filtered interval lists for all samples to generate the final interval list used to train the model for all samples. This step can be performed using the FilterIntervals function.

[0054] Step S11: Referring to step 3.6 of the technical solution, calculate the ploidy of the sample within the interval to construct a baseline model of interval ploidy. First, read the tsv format file of the coverage information of all samples calculated in step S9, use the principal component analysis (PCA) method to correct the coverage of different samples to ensure that the ploidy estimation is not affected by technical deviations and batch effects, and use the corrected coverage information as input. The specific principle is:

[0055] Assuming there are N samples and M genomic bins, the sequencing coverage data matrix is ​​represented as:

[0056] Among them, x i,j represents the sequencing coverage of sample i in interval j. In sequencing bias correction, the first few principal components usually represent batch effects or systematic biases. Therefore, when reconstructing the data, remove the first p principal components and retain the subsequent principal components: where X c It is a central matrix, which is obtained by subtracting the mean of each interval from the sequencing coverage of each interval, that is, X c =X-μ calculated, Z p is the component projected onto the first P principal components, V p is the corresponding principal component vector matrix.

[0057] Next, GMM is used to fit the coverage data to infer the ploidy state. The GMM probability density function (PDF) is as follows:

[0058]

[0059] Where x represents the observed coverage data point, K is the number of Gaussian distributions, and π k represents the weight of the kth distribution, satisfying and represents the kth Gaussian distribution with mean μ k and variance Based on the following likelihood function, the GMM optimal parameters are obtained by iteratively maximizing the likelihood estimation.

[0060]

[0061] By fitting the sequencing coverage distribution of the sample, μ k The closest distribution center is used to determine the sample's ploidy state. This step outputs the key parameters estimated during the model fitting process, which are stored and used for ploidy estimation in subsequent sample prediction processes. This step can be used in conjunction with gatk using the DetermineGermlineContigPloidy module to estimate ploidy parameters.

[0062] Step S12: Referring to step 3.7 of the technical solution, this step uses the files generated in steps S9-S11, inputs the ploidy baseline model parameters for the cohort, and the coverage files for all samples in the cohort to train a CNV model. Unsupervised learning methods are employed here, primarily using variational Bayesian inference (VBI) to model the coverage of different regions within the sample and infer potential copy number states. A hidden Markov model (HMM) is used to represent CNV state transition probabilities, learning the probability of CNV occurrence within a region and the variation in chromosome segment length. Coverage data are fitted using the HMM, and model parameters are optimized within a variational framework, ultimately resulting in a model that reflects the distribution of copy number states within the sample. Finally, a GMM is used to describe the coverage distribution and identify potentially polymorphic copy number regions. This step generates a file containing the variational parameters of the CNV model, the HMM state transition matrix, and copy number distribution information for each region. This constructed model effectively integrates sequencing data from multiple samples, accurately detecting rare and low-frequency CNVs, and minimizing the impact of technical noise on the results. In addition, the GermlineCNVCaller module in gatk can be used to train model parameters directly based on the queue input mode.

[0063] Step S13: Execute the above steps S2-S5, S9 and S11 on the WES sequencing data of the sample that needs to identify CNV to generate the sample coverage file and the ploidy assessment result file, which are input into the CNV identification model generated by step S12 to predict and infer the regions with abnormal copy number in the genome. The model parameters generated by S12 can also be called using the singleton mode (CASE mode) of the GermlineCNVCaller module in gatk for identifying samples that need to detect CNV based on WES sequencing data. The CNV identification results generated in this step can be saved as a text file in the following format:

[0064] chr1 12345 12500DEL

[0065] chr3 56789 58900DUP

[0066] Its contents represent the corresponding genomic interval deletion DEL and duplication DUP respectively.

[0067] After completing the above steps, you will obtain a model for identifying CNVs based on WES sequencing data and its related files that can be used to identify CNVs in the sample to be tested. The model and related files can be used to perform CNV detection on other samples to be tested. Figure 2-3 , Figure 2 The IGV tool was used to visualize reads and display abundance maps for the WES / WGS data of samples HG00383 and HG00179, respectively. HG00179 was used as a comparison, and it was observed that HG00383 had obvious read breakpoints and low abundance. The same observations were also made in the WGS data of HG00383, indicating that this sample had a copy number loss in the arrow region marked by the blue box (exon 5 of BPIFA3). Figure 3 The IGV tool was used to visualize reads and display abundance maps for the WES / WGS data of samples HG00099 and HG00383, respectively. By comparing the read abundance of HG00383 and adjacent positions, it can be observed that HG00099 has a significantly low abundance, and its related WGS data also have the same observation, both indicating that the sample has a copy number loss in the arrow region marked by the basket (exons 14-16 of SIGLEC1). This model can not only detect CNVs of single gene short fragments and single exons from samples of the 1000 Genomes, but also detect CNVs of single gene long fragments across exons. Therefore, the present invention will be able to implement single-gene-level CNV identification for patients with rare diseases based on WES sequencing data, which will improve the clinical utilization of WES data and further improve the diagnosis rate.

[0068] This embodiment provides a system, device, and medium for identifying single-gene copy number variations based on a whole-exome sequencing cohort. The computing device comprises a computational model system and a data storage medium. The computational model system integrates the aforementioned WES data processing, CNV model construction, and CNV model identification processes (i.e., steps S2-S13). The data storage medium stores the original WES data as well as intermediate result files and model files obtained through computation. The computational model system is functionally divided into a data reading module, a computational processing module, and a result output module. The computational processing module can be further divided into a data processing module, a model construction module, a training module, a validation and testing module, a model storage module, a model loading module, and a prediction and inference module. The details of the module integration procedures can be found in the corresponding steps above and will not be further described. The data storage medium comprises RAM and a solid-state drive (SSD) / hard disk (HDD). During the model and data loading process, data is read from the SSD / HDD. During the model training and prediction phases, RAM is used to store and transmit intermediate computation results. By rationally utilizing the storage medium, the computational model system improves data access speed, reduces I / O bottlenecks, and provides stable support for efficient training and prediction. To improve equipment operational efficiency, the computational model system will incorporate a resource management module, tightly integrating data storage media with computational tasks. This will create an efficient data flow and management system, ensuring smooth model training, inference, and analysis. Furthermore, a logging and monitoring module will be built to help monitor the training process in real time, recording key metrics such as loss and accuracy to facilitate analysis of model effectiveness.

[0069] The embodiments described above are merely some preferred embodiments of the present invention and are not intended to limit the present invention. Persons skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent substitution or equivalent transformation falls within the scope of protection of the present invention.

Claims

1. A method for identifying single-gene copy number variations based on whole-exome sequencing cohorts, characterized by: include: Step 1: Collect WES sequencing data of individual samples, perform quality control on the sample sequencing data, and screen out qualified WES sequencing data; Step 2: Preprocess the qualified WES sequencing data to obtain a file after each WES sequencing data is aligned with the human reference genome; Step 3: All the alignment files obtained in step 2 are input into the unsupervised learning model. The parameters are continuously adjusted through iterative optimization to make the model better fit the data distribution, obtain the optimal baseline model and related model parameters for CNV identification, and obtain the best model; Step 4: Perform the same data quality control as in step 1 on the WES sequencing data of the individual samples for CNV testing, then perform sequence alignment in step 2. Perform CNV identification based on the aligned files and the best model generated in step 3.

2. The method for identifying single-gene copy number variations based on whole-exome sequencing cohort according to claim 1, characterized in that: The WES sequencing data of individual samples in step 1 is determined according to the model requirements. If the model focuses more on universality, WES sequencing data of different sequencing backgrounds and more samples are selected. If the model focuses more on CNV identification accuracy, WES sequencing data of a single sequencing background is used to construct the model for CNV identification of WES sequencing data with the same sequencing background.

3. The method for identifying single-gene copy number variations based on whole-exome sequencing cohort according to claim 1, characterized in that: The pretreatment in step 2 specifically includes: Step 2.1: Remove the sequence adapters and perform base filtering on the qualified WES sequencing data to obtain the trimmed sequence file; Step 2.2: Align the sequence file obtained in step 2.1 to the human reference genome to obtain an intermediate file. The selected human reference genome version must remain consistent in subsequent analysis steps. Step 2.3: Perform the duplicate sequencing read length marking and base scoring correction steps on the intermediate files after the above alignment to obtain the final aligned files for each WES sequencing data.

4. The method for identifying single-gene copy number variations based on whole-exome sequencing cohort according to claim 1, characterized in that: The step 3 includes: Step 3.1: Create a list of CNV intervals to be identified based on the annotation files of the exon coordinates on the human reference genome provided by different institutions, or based on the exon coordinate position annotation files of the human reference genome; Step 3.2: Preprocessing the interval list, the preprocessing includes: adding padding space to the created interval list, and cutting the interval list; Step 3.3: Annotate the preprocessed interval sequences to improve the accuracy of interval CNV identification; Step 3.4: Based on the WES sequencing data alignment files for all samples obtained in step 2, use the preprocessed interval list to calculate the sequence read lengths for each sample in each interval and generate coverage information. The coverage is the depth of the sequencing reads covering the specified genomic interval, that is, the total number of aligned bases divided by the length of the target region; Step 3.5: Based on the coverage information of all samples generated in step 3.4, use the annotation information of the interval list in step 3.3 as a coverage covariate to correct the interval list. Based on the coverage distribution characteristics of the corrected samples, further filter the interval list to filter out the coverage abnormal intervals outside the confidence interval to form a new interval list. Step 3.6: For the new interval list, calculate the ploidy of the sample within the interval to build a baseline model of interval ploidy; Step 3.7: Use the queue mode, that is, input the information of all training samples at the same time, combine the sample coverage information obtained in step 3.4 and the baseline model information generated in step 3.6 as input features, use the unsupervised learning method to train the model, and obtain the CNV recognition model with the optimal parameters.

5. The method for identifying single-gene copy number variations based on whole-exome sequencing cohort according to claim 4, characterized in that: In step 3.3, the annotation includes the GC base content of the interval sequence and sequence alignment information, wherein the GC base content is the percentage of guanine G and cytosine C in the DNA sequence, and the sequence alignment rate refers to the degree to which the sequence can be uniquely mapped to the reference genome.

6. The method for identifying single-gene copy number variations based on whole-exome sequencing cohort according to claim 4, characterized in that: The filtering described in step 3.5 is specifically as follows: calculating the coverage of all exon target regions, statistically analyzing the distribution characteristics of the coverage, calculating the mean and standard deviation, filtering out regions outside the 95% confidence interval, and forming a new interval list.

7. The method for identifying single-gene copy number variations based on whole-exome sequencing cohort according to claim 4, characterized in that: Step 3.6 specifically includes: for the sequencing coverage information of each sample in the exon interval in the new interval list, use principal component analysis (PCA) to correct the coverage of different samples to ensure that the ploidy estimation is not affected by technical bias and batch effects, use Gaussian mixture model (GMM) to fit the coverage data, determine the most likely ploidy based on maximum likelihood estimation (MLE), and obtain the baseline model and its parameters.

8. A single-gene copy number variation identification system based on whole-exome sequencing cohort, characterized by: Used to implement the method according to any one of claims 1 to 7.

9. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions, wherein the instructions are used to implement the method according to any one of claims 1 to 7 when executed.

Images