Pediatric Rare Disease Whole Genome Sequencing Data Analysis System

Abstract

This invention relates to the field of bioinformatics analysis technology, specifically a whole-genome sequencing data analysis system for rare childhood diseases. It includes: a data preprocessing module for quality assessment and trimming of raw sequencing data to generate quality-controlled reads; an SNV / InDel analysis module for receiving the quality-controlled reads generated by the data preprocessing module, performing parallel alignment with dual reference genomes, sequencing platform-specific bias correction, and rare disease-specific optimized variant calling, and generating candidate SNV / InDel variants; an SV / CNV analysis module for receiving the quality-controlled reads generated by the data preprocessing module and generating candidate SV / CNV variants; and a variant annotation and filtering module for annotating and filtering candidate SNV / InDel and candidate SV / CNV variants generated by the SNV / InDel and SV / CNV analysis modules, in conjunction with a local frequency database. This invention significantly reduces the false positive rate and improves the overall accuracy and reliability of variant detection.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics analysis technology, specifically a whole-genome sequencing data analysis system for rare childhood diseases. Background Technology

[0002] In the field of whole-genome sequencing data analysis for rare childhood diseases, the analysis process needs to accurately detect rare variants related to the disease from massive amounts of raw data. Existing technical solutions usually adopt standardized bioinformatics analysis processes. When processing SNVs / InDels, these processes often use fixed filtering strategies. When processing SVs / CNVs, they often rely on a single detection algorithm or default parameter configurations.

[0003] This approach has significant limitations: specific biases and systematic artifacts introduced by different sequencing platforms, if not corrected, can lead to distorted analysis results; standard and stringent filtering criteria can easily misclassify low-frequency or newly emerging mutations, which are crucial for rare diseases, as noise and lose them; SV / CNV detection, due to its reliance on default parameters and the limitations of a single algorithm, generally suffers from high false positive rates, low recall rates, and a lack of reproducibility of results across different batches; the filtering strategy relies too heavily on international public databases and cannot effectively remove high-frequency benign variants specific to the local population, resulting in a large amount of background noise interference.

[0004] This serial, non-optimized processing mode results in insufficient sensitivity, accuracy, and stability of the analysis process. Therefore, how to systematically solve the problems of low-frequency mutation loss and unstable SV / CNV detection caused by platform bias, algorithm limitations, and coarse filtering strategies is a technical problem that urgently needs to be solved in this field.

[0005] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention discloses a whole-genome sequencing data analysis system for rare childhood diseases. Specifically, the technical solution of this invention includes:

[0007] The data preprocessing module is used to perform quality assessment and trimming on the raw sequencing data to generate quality-controlled reads;

[0008] The SNV / InDel analysis module is used to receive the quality-controlled reads generated by the data preprocessing module, perform parallel alignment of dual reference genomes, sequencing platform-specific bias correction, and rare disease-specific optimized variant calling to generate candidate SNV / InDel variants.

[0009] The SV / CNV analysis module is used to receive the quality-controlled read length generated by the data preprocessing module, perform hierarchical parameter calibration based on the parameter training library, and use multi-algorithm complementary detection and automated result integration to generate candidate SV / CNV variants.

[0010] The variant annotation and filtering module is used to annotate and filter candidate SNV / InDel variants generated by the SNV / InDel analysis module and candidate SV / CNV variants generated by the SV / CNV analysis module, in conjunction with a local frequency database.

[0011] Preferably, the SNV / InDel analysis module is specifically used for:

[0012] The quality-controlled reads were aligned to the reference genomes of hg19 and hg38, respectively.

[0013] By using precise coordinate mapping and cross-validation, the consistency of site information is ensured;

[0014] Perform sequencing platform-specific bias correction, which includes GC bias correction, Phred quality value unification, and local reassembly of homologous sequence regions;

[0015] Variance detection was performed using GATKHaplotypeCaller, and the filtering criteria were adjusted using a rare disease-specific optimization strategy.

[0016] Preferably, the rare disease-specific optimization strategy includes:

[0017] Adjust the filtering parameters and appropriately relax the filtering criteria for low-frequency mutations to achieve protection against low-frequency mutations;

[0018] By introducing a Bayesian statistical model, combining variation quality, coverage depth and chain bias information, the filtering threshold is dynamically adjusted.

[0019] By combining gene function annotation databases, variants located in known pathogenic genes or key functional domains are given candidate priority ranking.

[0020] Preferably, the SV / CNV analysis module performs hierarchical parameter calibration, specifically for:

[0021] Construct a parameter training library that includes international standard samples and locally validated positive samples;

[0022] Stratification is defined based on sequencing depth, variant length, and regional characteristics;

[0023] For each layer traversal parameter matrix, a performance metric is computed on the training library.

[0024] The optimal parameters are selected based on the preset objective function and stored in a parameter library for automatic use by the system when analyzing new samples.

[0025] Preferably, the SV / CNV analysis module employs complementary multi-algorithm detection, specifically used for:

[0026] Select complementary combinations of detection tools based on the type of signal source;

[0027] For short-to-medium fragment variations, a combination of tools based on SR / PE signals should be prioritized.

[0028] For large copy number variations, a tool combination based on RD / BAF signals should be preferred.

[0029] Preferably, the tool combination based on SR / PE signals includes Manta and Delly;

[0030] The tool suite based on RD / BAF signals includes CNVnator and ERDS.

[0031] Preferably, the SV / CNV analysis module performs automated result integration, specifically for:

[0032] By using genomic repetitive sequence databases and comparability thresholds, candidates within low-comparability regions or high-frequency repetitive elements are eliminated.

[0033] Convert the output of multiple tools into a unified VCF-compatible format;

[0034] The SURVIVOR tool was used to perform clustering and refinement based on preset overlap thresholds and layer distance parameters;

[0035] The breakpoint regions of overlapping candidates are re-aligned, and the boundaries are dynamically corrected using read length support number.

[0036] Preferably, the mutation annotation filtering module is specifically used for:

[0037] Common variants with frequencies higher than a preset threshold in the local frequency database will be filtered or marked as benign.

[0038] Compare local database frequencies with those of large international databases to identify local high-frequency or low-frequency variations.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. This invention innovatively employs a dual-reference genome parallel alignment (hg19 and hg38) using the SNV / InDel analysis module, combined with precise coordinate mapping and cross-validation, ensuring a high degree of consistency in site information. Simultaneously, the system performs sequencing platform-specific bias corrections, including GC bias correction and homologous sequence region reassembly, significantly reducing the false positive rate and improving the overall accuracy and reliability of variant detection.

[0041] 2. This invention addresses the needs of rare disease detection in children by employing a specific optimization strategy. By introducing a Bayesian statistical model and combining information such as coverage depth and chain bias, the filtering threshold is dynamically adjusted, and the criteria are appropriately relaxed to protect low-frequency mutations. Furthermore, by combining a functional annotation library to prioritize variants of known pathogenic genes, the detection rate and analysis efficiency of rare disease-related pathogenic variants are significantly improved.

[0042] 3. The SV / CNV analysis module of this invention addresses the complexity of detection by utilizing hierarchical parameter calibration. The system constructs a training library containing international standards and local validation samples, automatically selecting optimal parameters based on sequencing depth and regional characteristics. Simultaneously, a multi-algorithm complementary detection strategy is employed, utilizing the best tool combination for different signal sources and variant types, achieving automation and precision in detection.

[0043] 4. This invention effectively improves the quality of SV / CNV results through automated result integration and refinement. The SURVIVOR tool is used to cluster results from multiple algorithms and dynamically correct breakpoint boundaries, while simultaneously removing interference from high-frequency repetitive elements. The system incorporates a local frequency database for annotation filtering, effectively filtering common local variants and identifying unique high-frequency or low-frequency variants, thus enhancing the clinical applicability of the analysis results. Attached Figure Description

[0044] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0045] Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0047] Example 1:

[0048] Please see Figure 1 A system for analyzing and optimizing whole-genome sequencing data of rare childhood diseases, including:

[0049] The data preprocessing module is used to perform quality assessment and trimming on the raw sequencing data to generate quality-controlled reads;

[0050] The SNV / InDel analysis module receives the quality-controlled reads generated by the data preprocessing module, performs parallel alignment of dual reference genomes, corrects for sequencing platform-specific biases, and calls for rare disease-specific optimized variants to generate candidate SNV / InDel variants.

[0051] The SV / CNV analysis module receives the quality-controlled read length generated by the data preprocessing module, performs hierarchical parameter calibration based on the parameter training library, and uses complementary multi-algorithm detection and automated result integration to generate candidate SV / CNV variants.

[0052] The variant annotation and filtering module is used to annotate and filter candidate SNV / InDel variants generated by the SNV / InDel analysis module and candidate SV / CNV variants generated by the SV / CNV analysis module, in conjunction with the local frequency database.

[0053] In a specific embodiment of the present invention, the system is constructed as an automated bioinformatics analysis process and deployed in a high-performance computing cluster environment; through the collaborative work of four major modules, the system constructs a complete automated analysis closed loop from raw data to candidate variants.

[0054] The data preprocessing module is designed to ensure the quality of data entering subsequent analysis processes. This module receives raw WGS sequencing data from the sequencer as input and calls tools such as FastP to perform quality assessment, automatically removing adapter sequences, trimming low-quality bases, and filtering out excessively short reads. This module ultimately generates high-quality, quality-controlled reads as output, which serve as key data inputs and are distributed to two parallel core analysis modules.

[0055] The SNV / InDel analysis module is specifically designed to detect small fragment variants. Its core purpose is to identify SNVs / InDels associated with rare childhood diseases while addressing reference genome compatibility issues and platform bias. It receives quality-controlled reads generated by the data preprocessing module and performs parallel alignment of dual reference genomes, sequencing platform-specific bias correction, and variant recall strategies optimized for rare diseases. Its output is a list of candidate SNV / InDel variants.

[0056] The SV / CNV analysis module is responsible for detecting structural and copy number variations in large fragments. Its purpose is to address the limitations of single-algorithm detection and the difficulty of parameter tuning. It also receives reads after quality control and innovatively adopts hierarchical parameter calibration based on a parameter training library, as well as a strategy of multi-algorithm complementary detection and automated result integration to generate a candidate SV / CNV variant list. The parameter training library here is defined as a gold standard dataset for systematic optimization of algorithm parameters, and its construction source is dual: it includes both publicly available international standard samples and locally validated positive samples.

[0057] The variant annotation and filtering module serves as the final convergence point of the process. Its purpose is to refine the massive number of variants detected upstream and identify the most clinically relevant candidate variants. It receives candidate SNV / InDel variants and candidate SV / CNV variants generated by the two modules mentioned above, and uses a key local frequency database to annotate and filter these variants. The local frequency database has a special technical meaning in this invention. It is a dynamically updated dataset that characterizes the frequency spectrum of variants specific to the local population. Its data source is constructed by statistically analyzing massive, high-quality local healthy control group samples using WGS.

[0058] Through the collaborative work of the four modules mentioned above, this invention constructs a complete automated analysis closed loop from raw data to candidate variants. Compared with the serial processing and independent analysis process in existing technologies, the core innovation of this invention lies in the parallel and deep optimization at both the SNV / InDel and SV / CNV levels, and the use of a localized frequency database for convergence filtering at the output. This parallel optimization and local convergence design systematically solves the problems of low-frequency mutation loss and SV / CNV detection instability caused by platform bias, algorithm limitations, and stringent filtering strategies in existing technologies. This system significantly improves the sensitivity, accuracy, and reproducibility of WGS data analysis for rare childhood diseases.

[0059] Example 2

[0060] The SNV / InDel analysis module is specifically used for:

[0061] The quality-controlled reads were aligned to the reference genomes of hg19 and hg38, respectively.

[0062] By using precise coordinate mapping and cross-validation, the consistency of site information is ensured;

[0063] Perform sequencing platform-specific bias corrections, including GC bias correction, Phred quality value unification, and local reassembly of homologous sequence regions;

[0064] Variance detection was performed using GATKHaplotypeCaller, and the filtering criteria were adjusted using a rare disease-specific optimization strategy.

[0065] Rare disease-specific optimization strategies include:

[0066] Adjust the filtering parameters and appropriately relax the filtering criteria for low-frequency mutations to achieve protection against low-frequency mutations;

[0067] By introducing a Bayesian statistical model, combining variation quality, coverage depth and chain bias information, the filtering threshold is dynamically adjusted.

[0068] By combining gene function annotation databases, variants located in known pathogenic genes or key functional domains are given candidate priority rankings.

[0069] In this specific configuration of the embodiment, the internal working mechanism of the SNV / InDel analysis module is designed to systematically address compatibility issues between different reference genome versions and systematic errors introduced by the sequencing platform;

[0070] This module receives the quality-controlled reads and calls the alignment tool to initiate two independent alignment processes, aligning the quality-controlled reads to the reference genomes of hg19 and hg38 respectively. To resolve the potential coordinate confusion that may result from the dual-version alignment, the module then uses a precise coordinate mapping tool and cross-validation logic to compare and confirm the key site information in the alignment results of the two versions, ensuring the consistency of the site information.

[0071] After alignment, to eliminate systematic artifacts caused by different sequencers, the module immediately initiates sequencing platform-specific bias correction; this step is one of the key technologies of this invention; the bias correction specifically includes: GC bias correction, that is, standardizing the sequencing depth according to GC content; Phred quality value unification, that is, normalizing the quality value system of different platforms; for example, uniformly converting to the Sanger standard Phred+33 encoding format; and local reassembly of homologous sequence regions, that is, in known high repetition regions that are prone to mismatch, a local reassembly strategy is adopted to improve alignment accuracy;

[0072] The data stream that has completed the bias correction is then imported into the mutation detection step. In this embodiment, GATKHaplotypeCaller is used for mutation detection. The key to this solution is that this step adopts a rare disease-specific optimization strategy to adjust the filtering criteria. The purpose is to prevent the standard process from filtering out low-frequency mutations or new mutations that are crucial to childhood rare diseases as noise.

[0073] In this embodiment, the rare disease-specific optimization strategy is implemented through three synergistic sub-strategies:

[0074] Low-frequency mutation protection: Low-frequency mutation protection is achieved by adjusting filtering parameters, such as appropriately relaxing the lower limit of the allele frequency threshold and reducing the requirement for minimum support reads. Specifically, the allele frequency threshold used for system errors in the standard procedure, for example, 0.02, is relaxed to 0.01, and the minimum support reads are reduced from 5 to 3. These thresholds are set by those skilled in the art based on an empirical balance of the signal-to-noise ratio of new mutations, aiming to ensure that a low proportion of real mutations are not mistakenly filtered out as sequencing noise.

[0075] Dynamic threshold adjustment: The technical advantage of this strategy lies in abandoning fixed hard thresholds and instead introducing a Bayesian statistical model. This model can comprehensively consider multi-dimensional evidence such as variant quality, coverage depth, and chain bias, rather than relying on a single threshold, thereby dynamically adjusting the confidence score and filtering threshold of each variant, protecting genuine low-confidence variants while controlling false positives. As an implementation method, the core of this model is to calculate the posterior probability of a variant being true given data, D, including depth, quality, etc., and make a judgment based on this probability rather than a fixed threshold. For example, a Naive Bayes model can be used, and its calculation formula can be simplified as follows:

[0076]

[0077] in The prior probability of mutation, for example, set based on the known mutation rate. , It is a likelihood function obtained statistically from a parameter training library; the system will calculate this post-abortive probability and compare it with a preset probability threshold, such as 0.9, to dynamically adjust the filtering decision.

[0078] Candidate Priority Ranking: By combining gene function annotation databases, when a detected variant is located in a known pathogenic gene or key functional domain, the system automatically assigns it a candidate priority ranking. As one way to achieve this, a specific Priority=High tag can be added to the INFO field of the VCF file, so that downstream analysis software or human reviewers can immediately identify and prioritize these high-value candidates.

[0079] Parallel alignment and consistency verification of dual reference genomes ensure the compatibility and accuracy of the analysis results; platform-specific bias correction effectively reduces artifacts related to sequencing platforms; and the combined application of rare disease-specific optimization strategies can improve the sensitivity of capturing low-frequency mutations and new mutations without significantly increasing the false positive rate, which is crucial for exploring the etiology of rare childhood diseases.

[0080] Example 3

[0081] The SV / CNV analysis module performs hierarchical parameter calibration, specifically for:

[0082] Construct a parameter training library that includes international standard samples and locally validated positive samples;

[0083] Stratification is defined based on sequencing depth, variant length, and regional characteristics;

[0084] For each layer of traversal parameter matrix, calculate the performance metric on the training library;

[0085] The optimal parameters are selected based on the preset objective function and stored in a parameter library for automatic use by the system when analyzing new samples.

[0086] In this embodiment, a core innovation of the SV / CNV analysis module lies in its ability to perform hierarchical parameter calibration; its purpose is to address the problem of existing SV / CNV detection tools relying on default parameters, leading to high false positives and poor reproducibility; the calibration process of this invention is a systematic, data-driven optimization process:

[0087] Before the process starts, the system builds a parameter training library. The parameter training library is a gold standard dataset for optimizing algorithm parameters. Its source is dual: it includes both publicly available international standard samples and locally verified positive samples. This dual-source construction method ensures that parameter tuning is both internationally comparable and has the ability to calibrate for the specific biases of the local sequencing platform.

[0088] After constructing the training library, the system defines stratification based on sequencing depth, variant length, and regional characteristics. This stratification is a key optimization strategy, and its technical consideration is that the optimal detection parameters are drastically different for different genomic scenarios or different variant types. This invention defines refined analytical stratification based on sequencing depth, variant length, and regional characteristics.

[0089] For each layer, the system defines a parameter matrix that contains the key adjustable parameters of the tools used; the system iterates through these parameter combinations in the parameter training library and automatically calculates performance metrics.

[0090] After the evaluation is complete, the system selects the optimal parameters based on a preset objective function. In this embodiment, the preset objective function is designed as a composite metric that prioritizes recall while also considering precision, aiming to maximize the discovery of experimentally validated true variants. For example, the F-betascore can be used. The score, as the objective function, is defined as follows: ,in To give recall a higher weight; the selected optimal parameters are fixed into a parameter library so that the system can automatically call the corresponding optimal configuration according to the characteristics of the new sample when analyzing new samples;

[0091] By performing stratified parameter calibration, this invention transforms SV / CNV detection from a coarse-grained mode relying on default parameters to a precise mode based on stratified optimization. Calibration using a training library containing local positive samples improves the adaptability of parameter configuration to local sequencing platforms and population backgrounds. This mechanism of automatically selecting and solidifying optimal parameters ensures the stability and reproducibility of analysis results across different batches, improving the recall and precision of SV / CNV detection for local samples.

[0092] Example 4

[0093] The SV / CNV analysis module employs complementary multi-algorithm detection, specifically for:

[0094] Select complementary combinations of detection tools based on the type of signal source;

[0095] For short-to-medium fragment variations, a combination of tools based on SR / PE signals should be prioritized.

[0096] For large copy number variations, a combination of tools based on RD / BAF signals should be used first.

[0097] Tool suites based on SR / PE signals include Manta and Delly;

[0098] The tool suite based on RD / BAF signals includes CNVnator and ERDS;

[0099] In this embodiment, to address the problem that a single algorithm cannot adequately handle different variation scales, the SV / CNV analysis module further employs a multi-algorithm complementary detection strategy. The core design concept of this invention is that no single algorithm can perfectly detect all types of SV / CNV, and the signal source types upon which different algorithms rely have natural complementarity.

[0100] Signal source type refers to different types of evidence in WGS data that can be used to infer structural variations; this invention utilizes two key complementary signals:

[0101] SR / PE signal; this signal is highly sensitive to the breakpoint location, so it is crucial for the accurate localization of short and medium segment variations; the system preferentially uses a tool combination based on SR / PE signal, and in this embodiment, the tool combination preferably includes Manta and Delly; Manta is chosen because it is suitable for small and medium segments and can accurately locate the breakpoint, while Delly is chosen because it has high stability for medium and long segments;

[0102] RD / BAF signal; this signal reflects copy number changes in large fragments, thus having a high recall advantage for detecting large copy number variations; the system preferentially uses a tool combination based on RD / BAF signal, which in this embodiment preferably includes CNVnator and ERDS; CNVnator is chosen because it is suitable for detecting copy number variations >1kb, and ERDS is chosen because it can combine depth and isotropic frequency signals to improve the detection confidence of large CNVs;

[0103] This multi-algorithm complementary detection strategy, compared with the existing technology that relies on a single tool or a simple stacking of tools, is innovative in that it establishes a logical division of labor based on the physical characteristics of the signal source. The combination of Manta and Delly ensures the detection rate and breakpoint accuracy of short and medium segment mutations, while the combination of CNVnator and ERDS ensures high recall capability for large copy number mutations. This division of labor, combined with hierarchical parameter calibration, achieves stable and efficient detection of SV / CNV across the entire scale, effectively avoiding the detection blind spots of a single algorithm at the mutation scale.

[0104] Example 5

[0105] The SV / CNV analysis module performs automated result integration, specifically for:

[0106] By using genomic repetitive sequence databases and comparability thresholds, candidates within low-comparability regions or high-frequency repetitive elements are eliminated.

[0107] Convert the output of multiple tools into a unified VCF-compatible format;

[0108] The SURVIVOR tool was used to perform clustering and refinement based on preset overlap thresholds and layer distance parameters;

[0109] The breakpoint regions of overlapping candidates are re-aligned, and the boundaries are dynamically corrected using the read length support number.

[0110] For the multiple result sets generated by the aforementioned multi-algorithm detection, the SV / CNV analysis module further performs automated result integration. This step aims to address the issues of breakpoint offsets, inconsistent data types, and complex integration of detection results from different algorithms. This automated integration process is designed to be executed in multiple stages.

[0111] In the initial stage of the process, the system uses a genomic repetitive sequence database and a comparability threshold for preliminary filtering. The comparability threshold is a key quality control parameter, which in this embodiment is set to an average comparability of greater than 0.8 within a 100bp region on either side of the candidate variant breakpoint. The technical basis for setting this threshold is that regions below 0.8 are generally considered to be low-complexity or high-repetition regions, whose alignment signals are unreliable and are high-risk areas for false positives. This standard can effectively eliminate candidate variants located in low-comparability regions or high-frequency repetitive elements.

[0112] After filtering, the system converts the outputs from multiple tools such as Manta, Delly, and CNVnator into a VCF-compatible format and retains the quality labels, laying the foundation for subsequent clustering.

[0113] After format standardization, the system uses the SURVIVOR tool and local scripts to cluster and refine the variants in the VCF file. This step is based on preset overlap thresholds and layer distance parameters. In this embodiment, the overlap threshold is set to 70%, meaning that variants detected by two different tools are considered to be the same event only if their coordinate overlap area reaches more than 70%. This setting achieves a technical balance between merging and retaining new events. The layer distance parameter is a more refined control, which dynamically adjusts the allowable breakpoint distance tolerance based on the breakpoint accuracy characteristics of different tools. For example, for the Manta tool with high breakpoint accuracy, its distance tolerance can be set to 10bp; while for large CNV detection tools, its tolerance can be relaxed to 1kb.

[0114] In the final stage of cluster refinement, for the overlapping candidates that have passed the clustering, the system will extract the original read length of their breakpoint region and perform local re-alignment; the system uses the read length support number to dynamically correct its boundary and selects the coordinates with the most read length support as the final accurate breakpoint; this step ensures that the final output variation boundary is data-driven, rather than simply taking the average value.

[0115] Through this automated integration process, this invention solves the problems of redundancy and inconsistency in results caused by multi-algorithm detection; it effectively reduces false positives by eliminating duplicate candidate regions using a comparability threshold; and it improves the boundary accuracy and stability of SV / CNV by using the SURVIVOR tool for clustering and combining re-alignment and boundary correction of breakpoint regions, reducing missed detections or duplicate counts caused by breakpoint offsets, resulting in a higher quality and more reliable integrated variant set.

[0116] Example 6

[0117] The variant annotation filtering module is specifically used for:

[0118] Common variants with frequencies above a preset threshold in the local frequency database will be filtered or marked as benign.

[0119] Compare local database frequencies with those of large international databases to identify local high-frequency or low-frequency variations.

[0120] In this embodiment, the variant annotation filtering module is the last step in the entire analysis process; its purpose is to further screen out the variants that are most relevant to rare childhood diseases and are rare in the local population from the candidate variants generated by the SNV / InDel module and the SV / CNV module.

[0121] The core of this module is the use of a local frequency database; this database is a key feature that distinguishes this invention from existing technologies. The local frequency database is dynamically updated, and its data is constructed by performing WGS analysis on a massive amount of high-quality local healthy control group samples, processing them according to the aforementioned SNV / InDel and SV / CNV analysis procedures of this invention, and then statistically analyzing the allele frequencies of all variant sites. Its maintenance method involves periodically incorporating new local sample data for rolling updates to ensure the accuracy of the frequencies. For example, this update mechanism is set up as an automated batch processing task; after every 100 newly added ethically reviewed and quality-controlled local healthy control group samples are added to the database, a full recalculation of the frequencies of all variant sites in the database is automatically triggered.

[0122] The specific working mechanism of this module is as follows:

[0123] Local common variant filtering: The module receives a list of candidate variants and queries their frequencies in the local frequency database one by one. For common variants with frequencies higher than a preset threshold, the system will automatically filter them or mark them as benign. The preset threshold here is key to screening rare disease variants. The specific value of this threshold is not fixed, but can be set and dynamically adjusted based on the genetic pattern of the disease under study and the genetic background of the local population, through a combination of statistical analysis and expert experience.

[0124] As a specific implementation method, this statistical analysis is defined as follows: the threshold is set to the 99.5th percentile of the frequency of all autosomal variants in the local frequency database, for example, 0.01 or 1%, to ensure that only clearly common (>1%) variants in the local population are filtered out, thereby balancing filtering efficiency and retention of pathogenic variants; aiming to achieve optimal filtering performance.

[0125] Local specificity identification: The module also compares the frequency of the variant in local databases with the frequency in large international databases; this comparison aims to identify high-frequency or low-frequency variants that are specific to the local population, providing key local population context information for subsequent clinical interpretation.

[0126] By constructing a localized frequency baseline, the technical challenge of filtering out local high-frequency benign variants that cannot be filtered by directly using international databases was addressed. This improved the signal-to-noise ratio, effectively filtering out a large number of background noise variants that are common in the local population and unrelated to the target rare disease. This allowed analysts to focus more on variants that are truly rare and more likely to cause disease in the local population, improving the adaptability of the workflow to the local population background and the efficiency of discovering rare disease etiological variants.

[0127] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A system for analyzing and optimizing whole-genome sequencing data of rare childhood diseases, characterized in that: include: The data preprocessing module is used to perform quality assessment and trimming on the raw sequencing data to generate quality-controlled reads; The SNV / InDel analysis module is used to receive the quality-controlled reads generated by the data preprocessing module, perform parallel alignment of dual reference genomes, sequencing platform-specific bias correction, and rare disease-specific optimized variant calling to generate candidate SNV / InDel variants. The SV / CNV analysis module is used to receive the quality-controlled read length generated by the data preprocessing module, perform hierarchical parameter calibration based on the parameter training library, and use multi-algorithm complementary detection and automated result integration to generate candidate SV / CNV variants. The variant annotation and filtering module is used to annotate and filter candidate SNV / InDel variants generated by the SNV / InDel analysis module and candidate SV / CNV variants generated by the SV / CNV analysis module, in conjunction with a local frequency database.

2. The whole-genome sequencing data analysis and optimization system for rare childhood diseases according to claim 1, characterized in that, The SNV / InDel analysis module is specifically used for: The quality-controlled reads were aligned to the reference genomes of hg19 and hg38, respectively. By using precise coordinate mapping and cross-validation, the consistency of site information is ensured; Perform sequencing platform-specific bias correction, which includes GC bias correction, Phred quality value unification, and local reassembly of homologous sequence regions; Variance detection was performed using GATKHaplotypeCaller, and the filtering criteria were adjusted using a rare disease-specific optimization strategy.

3. The whole-genome sequencing data analysis and optimization system for rare childhood diseases according to claim 2, characterized in that, The rare disease-specific optimization strategy includes: Adjust the filtering parameters and appropriately relax the filtering criteria for low-frequency mutations to achieve protection against low-frequency mutations; By introducing a Bayesian statistical model, combining variation quality, coverage depth and chain bias information, the filtering threshold is dynamically adjusted. By combining gene function annotation databases, variants located in known pathogenic genes or key functional domains are given candidate priority ranking.

4. The whole-genome sequencing data analysis and optimization system for rare childhood diseases according to claim 1, characterized in that, The SV / CNV analysis module performs hierarchical parameter calibration, specifically for: Construct a parameter training library that includes international standard samples and locally validated positive samples; Stratification is defined based on sequencing depth, variant length, and regional characteristics; For each layer traversal parameter matrix, a performance metric is computed on the training library. The optimal parameters are selected based on the preset objective function and stored in a parameter library for automatic use by the system when analyzing new samples.

5. The whole-genome sequencing data analysis and optimization system for rare childhood diseases according to claim 1, characterized in that, The SV / CNV analysis module employs complementary multi-algorithm detection, specifically for: Select complementary combinations of detection tools based on the type of signal source; For short-to-medium fragment variations, a combination of tools based on SR / PE signals should be prioritized. For large copy number variations, a tool combination based on RD / BAF signals should be preferred.

6. The whole-genome sequencing data analysis and optimization system for rare childhood diseases according to claim 5, characterized in that, The tool suite based on SR / PE signals includes Manta and Delly; The tool suite based on RD / BAF signals includes CNVnator and ERDS.

7. The whole-genome sequencing data analysis and optimization system for rare childhood diseases according to claim 1, characterized in that, The SV / CNV analysis module performs automated result integration, specifically for: By using genomic repetitive sequence databases and comparability thresholds, candidates within low-comparability regions or high-frequency repetitive elements are eliminated. Convert the output of multiple tools into a unified VCF-compatible format; The SURVIVOR tool was used to perform clustering and refinement based on preset overlap thresholds and layer distance parameters; The breakpoint regions of overlapping candidates are re-aligned, and the boundaries are dynamically corrected using read length support number.

8. The whole-genome sequencing data analysis and optimization system for rare childhood diseases according to claim 1, characterized in that, The mutation annotation filtering module is specifically used for: Common variants with frequencies higher than a preset threshold in the local frequency database will be filtered or marked as benign. Compare local database frequencies with those of large international databases to identify local high-frequency or low-frequency variations.

Images