Genomic variation quality control detection methods and systems

Abstract

This invention discloses a method and system for quality control detection of genomic variations, belonging to the field of bioinformatics technology. The method includes: constructing a genomic identifiability grid feature database; calculating the observational uniqueness alignment rate of each grid unit and statistically analyzing the average sequencing depth corresponding to each genomic grid unit; obtaining the genotype quality score of the grid unit as the variation quality information of the grid unit; and constructing a risk map of variation detection reliability. This invention is mainly used for the reliability detection and evaluation of genomic variations based on sequence identifiability.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics technology, specifically to a method and system for quality control detection of genomic variations based on sequence identifiability. Background Technology

[0002] The fields of bioinformatics, genomics data processing, and high-throughput sequencing analysis are intertwined in practice. Reliability assessment techniques for genomic variant detection based on sequence identification, as well as related quality control methods and systems, are rapidly evolving and iterating, with new technologies constantly emerging. With the rapid development of high-throughput sequencing technology, whole-genome sequencing (WGS) has become a crucial technological foundation for genetic disease diagnosis, precision oncology, and population genetics research. By performing large-scale sequencing on biological samples and combining it with bioinformatics analysis workflows, the detection and analysis of various genomic variants, such as single nucleotide variants (SNVs), insertions / deletions (InDels), and structural variations (SVs), can be achieved. In existing technological systems, genomic variant detection typically includes steps such as sequencing data acquisition, sequence alignment, variant identification, and quality control. Among these, quality control is the key factor determining the reliability of variant detection results. Current mainstream variant detection workflows primarily rely on statistical indicators such as sequencing depth, genotype quality (GQ), variant quality (QUAL), and mapping quality (MQ) to evaluate the reliability of variant results. It is generally believed that with increased sequencing depth and improved data coverage, the accuracy and stability of variant detection results can be continuously improved; therefore, increasing sequencing depth has become one of the main methods for improving detection quality in the industry. However, the aforementioned quality evaluation system and corresponding detection techniques are essentially based on statistical repeated observations, focusing primarily on whether the amount of sequencing data is sufficient, while giving less consideration to whether the target region possesses the prerequisites for accurate identification and unique localization. In reality, the human genome is not a uniform and simple linear sequence, but contains a large number of complex repetitive structures and highly homologous regions. For example, the human genome widely contains complex structures such as segmental duplication, paralogous gene families, low-complexity sequences, multicopy repetitive regions, and pseudogenes. These regions often exhibit extremely high sequence similarity, making it difficult to effectively distinguish sequencing fragments from different genomic locations at the sequence level. For currently widely used short-read sequencing technologies, since the length of a single sequencing fragment is typically only tens to hundreds of bases, when the sequencing read originates from the aforementioned complex regions, it often matches multiple genomic locations simultaneously, resulting in multiple-mapping.To obtain the final alignment results, existing alignment software typically employs strategies such as random allocation, retention of multiple positions, reduction of alignment quality scores, or direct discarding to handle such reads. While these strategies maintain the normal operation of the analysis workflow, they do not fundamentally solve the problem of the lack of unique location information in the target region. More importantly, in highly homologous regions, even with continuous increases in sequencing depth, the newly acquired sequencing reads may still carry the same uncertain location information, thus offering extremely limited improvement in the ability to detect true variants. In other words, the problem faced by some regions is not insufficient sequencing data, but rather the lack of sufficient discriminative features in their own sequence structure, preventing the sequencing system from obtaining information that uniquely points to the true source location. This limitation stems from the underlying sequence structure of the genome itself, rather than the performance of the sequencing platform or the data scale, and is therefore relatively stable and difficult to eliminate by increasing sequencing depth. Existing technologies typically address this problem using empirical blacklist region filtering, repetitive sequence annotation filtering, or manual review. However, these methods mostly rely on post-hoc experience summaries, lack unified quantitative evaluation standards, and are difficult to automate reliability assessment and detection across the entire genome. Furthermore, existing quality control systems primarily evaluate the variant detection results themselves, lacking a technical means to quantitatively describe whether a target region possesses unique observability, starting from the source of sequencing information. Between different regions of the genome, there exists an inherent property determined by the sequence itself: the ability of a sequencing fragment in a certain region to be accurately and unambiguously mapped back to its true origin location. This property is mainly influenced by factors such as sequence repetition, homology complexity, and local sequence composition, and is not directly related to the amount of data obtained in a single sequencing experiment. For regions with high uniqueness, increasing sequencing depth usually effectively improves variant detection quality; however, for regions with extremely low uniqueness, even with higher depth data, variant detection results may remain unreliable for a long time. However, to date, there is a lack of technical means and corresponding schemes to systematically characterize this property at the whole-genome scale, and also a reliability assessment system and detection technology that directly correlates this property with variant detection quality. Existing technologies have not yet established a standardized database for the spatial structural features of the genome, making it impossible to automatically identify and classify the risks of detecting complex regions, and also making it difficult to provide clear guidance for subsequent long-read sequencing validation, complex region variant reconstruction, and the construction of high-confidence truth sets.

[0003] Among some related prior art, specifically referencing the U.S. patent application US2025 / 0191682A1 (published June 12, 2025) and U.S. patent application 18 / 972,558 (filed December 6, 2024), which describes a short-read and long-read integrated sequencing technology for genomic variant detection, the coverage modulation engine selects short-read and long-read sequencing coverage based on the sequencing objectives of the nucleic acid sequencing event and at least one constraint. A sequencing protocol generator determines a short-read sequencing protocol based on the selected short-read sequencing coverage and a long-read sequencing protocol based on the selected long-read sequencing coverage. An integrated variant invocation module outputs variant invocation results based on the short-read sequencing data generated through the short-read sequencing protocol and the long-read sequencing data generated through the long-read sequencing protocol. The technology disclosed in this US patent application (US2025 / 0191682A1) also fails to adequately address the problems previously mentioned in this invention. While the patent application proposes an integrated scheme of short-read + long-read hybrid sequencing, it still has significant limitations in terms of fundamentality, systematicity, and universality, and cannot comprehensively address the related issues. It fails to establish a unified quantitative framework for "unique observability" from the outset. The patent application primarily alleviates multiple alignment problems by using long-read-assisted enhancement of the reference genome and hybrid alignment, but it does not propose a unified technical solution or standardized database that is independent of specific sequencing experiments and can systematically characterize the inherent unique observability / mappability of each region at the whole-genome scale. Furthermore, it does not disclose or provide a pre-calculated, standardized, and tiered system for evaluating genomic structural features, such as steps or methods for quantitative scoring based on sequence repetition and homology complexity, thus failing to achieve the goal of automatically identifying risks in complex regions beforehand. Essentially, this patent application still relies on "increasing data volume" (long-read coverage) rather than fundamentally solving structural limitations. The core of this patent application is Coverage Modulation (intelligent modulation of short-read + long-read coverage) and long-read enhanced reference. While this is smarter than simply increasing short-read depth, it still falls under the traditional approach of compensating with more / better data. For extremely complex regions with very low uniqueness, even using long reads requires sufficient coverage for effective resolution, and long reads themselves still have limitations in cost, throughput, and error rate. This patent application does not eliminate the fundamental limitations of structural determinism, but at best can only offer partial mitigation. Furthermore, this patent application lacks a truly independent and automated reliability assessment system. Existing problems emphasize the need for quantitative description of regional observability from the source of sequencing information and its direct correlation with variant detection quality.This patent application focuses more on optimizing downstream variant retrieval processes, such as hybrid alignment and refining algorithms, rather than establishing a standardized, platform-independent upstream reliability assessment method. It fails to address issues such as blacklist reliance on experience and the lack of unified standards, and it cannot directly guide subsequent long-read validation and truth set construction. The patent application lacks practical applicability, requiring simultaneous short-read and long-read hybrid sequencing, which significantly increases experimental costs, complexity, and time. For clinical / research scenarios heavily reliant on pure short-read WGS, the implementation threshold is high, making it unsuitable as a lightweight, universal upgrade solution. In a sense, this patent application resembles a specific hybrid sequencing optimization technology rather than a general-purpose tool and method for addressing genomic spatial structure characteristics. While the patent application offers some innovation in hybrid data integration and cost optimization, and can partially improve detection performance in complex regions, it fails to provide a systematic, fundamental, and standardized solution to the problems, especially lacking a unified quantitative assessment framework with genome-wide unique localization capabilities and a reliability management system and methods independent of sequencing depth.

[0004] In some relevant prior art, specifically referring to Chinese patent application titled "A Method and System for Genome Variation Detection Based on Few-Sample Learning," with authorization announcement number CN119993261B, authorization announcement date September 9, 2025, and application number 202510076029.6, application date January 17, 2025, a method and system for genome variation detection based on few-sample learning is disclosed. The genome variation detection model constructed by this method integrates an adaptive memory module, a supervised contrastive learning module, and an orthogonal transfer module. The adaptive memory module enhances the effective preservation and expression of contextual features by dynamically storing and updating key information; the supervised contrastive learning module optimizes feature representation using a contrastive learning strategy, improving the semantic distinction ability between entities and relationships; and the orthogonal transfer module achieves efficient transfer and adaptation of cross-task knowledge by introducing orthogonal constraints, significantly enhancing the model's generalization ability. The aforementioned methods and systems, through collaborative optimization of entity recognition tasks, achieve outstanding performance, particularly demonstrating excellent detection accuracy and robustness in extremely limited sample settings such as 6-Shot-1-Way, providing an efficient and accurate solution for genomic variant detection. However, the technology disclosed in Chinese patent CN119993261B does not adequately address the problems previously mentioned in this invention. The core of this patent is a few-shot learning-based NLP text mining method used to extract genomic variant entities from text descriptions in biomedical literature and databases (similar to named entity recognition tasks). The disclosed technology has weak relevance to high-throughput sequencing (WGS) raw data processing, sequence alignment, and variant detection. It targets variant entity recognition, using BioBERT, adaptive memory, supervised contrastive learning, orthogonal transfer learning, essentially resembling information extraction and text processing techniques. This patent does not focus on short / long read sequencing data, sequence alignment algorithms, coverage modulation, or structural variant analysis, and therefore cannot directly impact the WGS data processing workflow from a certain perspective. This patent also fails to fully disclose genomic structural issues such as sequence alignment, multiple mapping, segmental duplication, and pseudogenes, and does not propose solutions for enhancing the reference genome, mixed alignment, or mappability quantification. It only optimizes the generalization ability of entity recognition at the text level, and in limited sample scenarios, it cannot address the limitation that sequencing reads cannot be uniquely mapped back to real genomic locations. The patent focuses on loss function optimization, contrastive learning, and orthogonal constraints in model training to improve the F1 score of text entity recognition, but it does not disclose any system or method for quantitatively assessing the inherent observability of regions starting from the sequencing source. The patent does not provide any genome-wide structural feature quantification tools, blacklist alternatives, or automated reliability assessment frameworks, and it cannot guide subsequent long-read validation or ground truth set construction.In variant detection in highly homologous and complex regions, it neither reduces multi-mapping nor compensates for structural defects in sequencing data by increasing text training data. This patent fails to provide solutions to constraints determined by sequence structure, nor does it establish unified quantitative evaluation systems and methods; therefore, it cannot solve the various core challenges mentioned in this invention for WGS variant detection.

[0005] Therefore, the field covered by this invention requires the development of new quality control testing and evaluation methods and systems to overcome related defects and shortcomings. Meanwhile, the aforementioned prior art documents are incorporated herein by reference in their entirety. Summary of the Invention

[0006] To address the shortcomings of the prior art, the present invention provides a method and system for quality control detection of genomic variations, thereby solving at least one of the aforementioned technical problems.

[0007] In a first aspect, the present invention provides a method for quality control detection of genomic variations.

[0008] In one embodiment of the first aspect of the present invention, the method includes: acquiring a reference genome sequence; dividing the reference genome sequence into continuous and non-overlapping grid units along chromosome coordinate axes according to a preset baseline length; assigning a unique physical coordinate identifier to each grid unit, wherein the physical coordinate identifier is stored in association with chromosome coordinate information, thereby constructing a genome identifiability grid feature database; acquiring sequencing data alignment files of the target sample; calculating the observed unique alignment rate of each grid unit using the grid unit as a statistical unit, wherein the observed unique alignment rate is the ratio of the number of unique mapped reads falling within the grid unit and satisfying preset alignment conditions to the total number of reads allocated to the grid unit; simultaneously, calculating the average sequencing depth corresponding to each genome grid unit; extracting the initial gene variant file for the target sample output by the upstream variant detection tool; calculating and fusing the genotype quality score of each grid unit using a length-weighted average algorithm as the variant quality information of the grid unit; constructing a risk map characterizing the variant detection reliability of each grid unit based on the observed unique alignment rate, the average sequencing depth, and the variant quality information; and performing quality control processing on the variant detection results based on the risk map.

[0009] In another embodiment of the first aspect of the present invention, before dividing the reference genome sequence into continuous and non-overlapping grid units along the chromosome coordinate axis according to a preset baseline length, the method further includes: scanning the reference genome sequence, identifying and extracting effective sequencing intervals with continuous non-N bases and a length greater than a preset physical threshold, and removing unknown regions containing N bases; and performing the grid unit division only within the effective sequencing intervals.

[0010] In yet another embodiment of the first aspect of the invention, the observation uniqueness ratio o-UMR is calculated according to the following formula:

[0011] ,

[0012] R unique ,i represents the number of unique mapped reads within the i-th genome grid cell;

[0013] R total ,i represents the total number of reads assigned to the i-th genomic grid unit.

[0014] In another embodiment of the first aspect of the present invention, the observed unique alignment rate (o-UMR) and the average sequencing depth are used as two independent evaluation dimensions, and the genotype quality score of the corresponding grid unit is represented by a color gradient, thereby forming a whole-genome reliability risk distribution map; each grid unit corresponds to a discrete observation point in the map; each grid unit is divided into: high identifiability region; medium identifiability region; and low identifiability region according to a preset observed unique alignment rate threshold;

[0015] When o-UMR ≥ 0.75, it is defined as a highly identifiable region;

[0016] When 0.25 ≦ o-UMR < 0.75, it is defined as a region of moderate identifiability;

[0017] When o-UMR < 0.25, it is defined as a low identifiability region.

[0018] In yet another embodiment of the first aspect of the present invention, risk warnings are automatically marked for each point;

[0019] Summarize the genomic coordinate intervals corresponding to low identifiability regions; generate a targeted rescue recommendation list containing the genomic coordinate interval information to guide subsequent validation sequencing, resequencing analysis, or cross-sequencing platform validation analysis.

[0020] In another embodiment of the first aspect of the present invention, a genomic grid cell representing a low-identifiability region is used to perform a long-read rescue process using long-read sequencing data for targeted validation and quality reconstruction, comprising the following steps: generating a targeted rescue list for low-identifiability regions; extracting coordinate information of all grid cells marked as low-identifiability regions based on the generated quality control results; for grid cells containing clinically relevant genes, drug metabolism genes, genetic disease-related loci, or other target regions, the system automatically summarizes the corresponding chromosome coordinate intervals and generates a standardized BED format targeted rescue recommendation list; the targeted rescue recommendation list can be used as an input file for subsequent long-read sequencing validation or targeted resequencing analysis; long-read sequencing data acquisition... Retrieve; Obtain long-read sequencing data corresponding to the target sample; Perform alignment and variant detection analysis on the long-read sequencing data to generate corresponding variant result files, including gVCF files; Target reanalysis of low-identifiability regions; The system reads the generated BED format target rescue recommendation list and extracts alignment records and variant records covering the corresponding genomic coordinate regions from the long-read sequencing results; Single-base resolution quality reconstruction; Call the single-base resolution quality reconstruction algorithm to map the quality scores in the long-read variant files to the single-base precision matrix, so that each nucleotide position corresponds to an independent quality score; At nucleotide positions j where sequence blocks overlap, perform a single-base-level mathematical averaging operation to obtain the fused single-base quality score GQ. base,j The overall long read variation quality score (GQ) of the reconstructed grid cell is calculated using the following formula:

[0021] ;

[0022] Quality comparison analysis before and after rescue of different identifiable regions: Extract the sets of genomic grid units corresponding to the high identifiable region Q1, medium identifiable region Q2 and low identifiable region Q3 respectively; For each grid unit, respectively calculate: (1) the genotype quality score GQShort obtained from short read sequencing data; (2) the reconstructed genotype quality score GQLong obtained from the long read sequencing quality reconstruction process; The system uses grid units as statistical objects to group and compare GQShort and GQLong in the high identifiable region Q1, medium identifiable region Q2 and low identifiable region Q3, and uses non-parametric statistical test methods to evaluate the difference between the two groups of quality score distributions; Output the reconstruction results; The reconstructed variant quality score is correlated and compared with the short read sequencing results, and the corresponding reconstruction result file is generated; For variant sites that have been fully supported after long read sequencing verification, the reconstruction results are used to correct, supplement or replace the short read sequencing analysis results.

[0023] Secondly, the present invention provides a genomic variation quality control detection system.

[0024] In one embodiment of the second aspect of the present invention, the system includes: a database management unit for storing and loading a genome identifiability grid feature database, the database comprising continuous and non-overlapping grid cells segmented along chromosome coordinate axes at a preset baseline length and their corresponding physical coordinate identifiers; an observation unique alignment rate calculation unit for calculating the observation unique alignment rate of each grid cell in the target sample sequencing data based on the grid cells; a sequencing depth statistics unit for calculating the average sequencing depth of each grid cell; a quality parameter extraction unit for extracting variant quality information of each grid cell in the variant recall results; a risk map generation unit for constructing a whole-genome variant detection reliability risk map based on the observation unique alignment rate, the average sequencing depth, and the variant quality information; and a quality control unit for performing automatic labeling, downgrading, filtering, or rescue analysis on the variant detection results based on the risk map.

[0025] In another embodiment of the second aspect of the present invention, the invention further includes: a target rescue list generation unit for summarizing genomic coordinate intervals of low-identifiability regions and generating a standardized target rescue list; a long read data interface unit for acquiring long read sequencing variant data; a quality reconstruction unit for mapping long read variant quality scores to a single-base precision matrix and calculating the reconstruction quality scores of grid cells; and a result fusion unit for correcting, supplementing, or replacing short read variant detection results based on the reconstruction quality scores.

[0026] Compared with existing technologies, the outstanding advantages of this invention are as follows: By constructing a genome sequence identifiability grid feature database, this invention divides the entire genome into continuous and non-overlapping standardized grid units, and uses the observed unique alignment rate (o-UMR) as the core quantitative indicator, achieving a quantitative description of the identifiability of sequences in different regions of the genome. This indicator is determined by the underlying sequence structure of the genome, independent of sequencing depth, and can reflect whether the target region has the prerequisite for being uniquely observed from the information source, overcoming the shortcomings of existing technologies that rely solely on statistical repeated observation indicators. Based on the observed unique alignment rate, this invention divides genome regions into three levels of identifiability: high, medium, and low, and constructs a three-dimensional risk map containing observed unique alignment rate, average sequencing depth, and variant quality information. For variant sites in low-identifiability regions, risk labeling, quality downgrading, or result filtering can be automatically performed, achieving automated quality control across the entire genome and avoiding the limitations of traditional methods that rely on ex-post experience summaries and manual review. This invention automatically generates a targeted rescue recommendation list to guide subsequent long-read sequencing validation analysis and utilizes a single-base resolution quality reconstruction algorithm to supplement and reconstruct the quality of variant detection results in low-identifiability regions. Experiments demonstrate that in low-identifiability regions, long-read sequencing and single-base resolution quality reconstruction significantly improve overall genotype quality scores, with the quality distribution clearly shifting towards higher quality intervals, approaching the quality level corresponding to high-identifiability regions. This invention establishes a complete technical chain from genome grid unit division, observational unique alignment rate calculation, risk map construction to targeted rescue validation, providing a standardized data foundation and traceable quality control procedures for constructing high-confidence variant truth sets, possessing significant clinical application value and research promotion significance. This invention directly quantifies the sequence identifiability of genomic regions, establishing a reliability assessment system oriented towards whole-genome spatial structure, thereby achieving automatic identification of detection risks in complex regions, hierarchical quality control of variant results, and cross-sequencing platform data collaborative reconstruction, improving the authenticity and reliability of genomic variant detection results. Attached Figure Description

[0027] Figure 1 This invention provides a risk map for characterizing the reliability of variant detection;

[0028] Figure 2 This is a coverage-quality response curve diagram of region Q1 and region Q3 of the present invention;

[0029] Figure 3 This is a graph showing the quality improvement effect of long-read sequencing quality reconstruction in different identifiable regions according to the present invention;

[0030] Figure 4This invention presents a risk map of the reliability of characterization variant detection for long-read sequencing to rescue reconstructed regions with low identifiability.

[0031] Figure 5 This is a simplified reference diagram illustrating the construction of the genome identifiability grid cell feature database and the variant quality control interception process based on observational unique alignment rate (o-UMR) of the present invention. Detailed Implementation

[0032] The present invention will now be described in detail with reference to the accompanying drawings and embodiments:

[0033] Glossary: ​​Short-read sequencing (SRS): A sequencing technology based on second-generation high-throughput sequencing platforms that fragments nucleic acid into shorter sequences (typically tens to hundreds of bases) for massive parallel reading. Due to read length limitations, this technology is easily affected by interference when dealing with complex physical configurations containing highly homologous gene clusters or segmental repetitive sequences. Third-generation sequencing (TGS): A technology based on third-generation sequencing technologies such as single-molecule real-time sequencing (e.g., PacBio HiFi) or nanopore sequencing, capable of continuously reading nucleic acid fragments of thousands to tens of thousands of bases. Utilizing its advantage of long-fragment structural crossing, it traverses complex regions of the genome (such as repetitive sequences, structural variations, or highly polymorphic regions), thereby achieving the physical reconstruction of the true values ​​of variations in complex regions. Genotype Quality (GQ): A statistical confidence metric for the correctness of the genotype determined by the algorithm at a specific locus during bioinformatics variant calling. Sequence Identifiability: In the process of genome sequencing and alignment, the inherent spatial property of a nucleotide fragment originating from a specific genomic coordinate can overcome interference from physical structures such as paralogous homologs and repetitive sequences within the genome, and be clearly, unambiguously, and uniquely mapped back to its true physical origin coordinates. Its size is determined by the physical configuration of the basal genome and is independent of sequencing depth. Genomic Bins: Based on a specific reference genome sequence, the continuous and valid reference genome sequence after removing unknown N bases is divided along the coordinate axes into continuous, equidistant, and non-overlapping discrete intervals of a fixed physical length (e.g., 2000 bp). Observed Unique Mapping Rate (o-UMR): In actual sequencing experiments targeting a target sample, the number of uniquely mapped fragments that actually fall within a specific grid cell and meet preset alignment conditions (e.g., correct end-to-end pairing and alignment quality score higher than a set threshold) is the ratio to the total number of fragments assigned to that grid cell (including uniquely mapped fragments and multiple-mapped fragments). This metric reflects the measured physical identifiability of the region under a specific sequencing platform. The Identifiability Grid Feature Database is a structured database constructed based on a reference genome sequence, containing multiple genomic grid units and their corresponding physical coordinates. This database is used for unified grid indexing, regional localization, and subsequent identifiability analysis of sequencing data, providing a spatial coordinate basis for assessing the reliability of variant detection.Mean Depth: The statistical average of the sequencing coverage depth of all nucleotide sites within a specific genomic grid unit, used to characterize the sequencing coverage of that region. Variant Detection Reliability Risk Map: A two-dimensional, three-dimensional, or multi-dimensional association model constructed based on observed unique alignment rate, sequencing depth, and variant quality parameters, used to characterize the reliability level of variant detection in different genomic regions and provide a basis for subsequent quality control decisions. High-Identifiability Region: A genomic region where the observed unique alignment rate is higher than a preset threshold. In this region, most sequencing reads can be uniquely located, and the variant detection results have high reliability. Intermediate-Identifiability Region: A genomic region where the observed unique alignment rate is within a preset intermediate range. This region exhibits a certain degree of multiple mapping, and the variant detection results need to be comprehensively judged in conjunction with other quality parameters. Low-Identifiability Region: A genomic region where the observed unique alignment rate is lower than a preset threshold. This region typically contains highly homologous sequences, repetitive sequences, or complex structural regions, making it difficult for short-read sequencing reads to achieve stable and unique localization. Therefore, the results of variant detection in this region have high uncertainty. The Targeted Rescue Recommendation List refers to a set of target genome coordinates automatically generated based on low-identifiability regions. This list is used to guide subsequent long-read sequencing validation, resequencing analysis, or cross-sequencing platform validation analysis to supplement and reconstruct the quality of variant information in the corresponding regions.

[0034] This invention generally or exemplaryly relates to methods and steps for quantitatively assessing the reliability of variant detection in different regions of the genome by, for example, constructing an Identifiability Grid Feature Database, combined with Observed Unique Mapping Rate (o-UMR), sequencing depth, and variant quality parameters. It further includes methods, systems, and others for risk map construction, automatic identification of low-reliability regions, quality control of variant results, and quality reconstruction and validation analysis based on long-read sequencing data. Unless specifically described in text, flowcharts in this invention are generally or exemplaryly understood to represent information, data, or control from top to bottom and from left to right.

[0035] See Figure 5This invention provides a feasible embodiment 1 of the genome identifiability grid cell feature database construction and variation quality control interception process based on observed unique alignment rate (o-UMR). This embodiment details the construction process of an identifiability grid cell feature database based on a specific reference genome (such as the internationally accepted human standard reference genome GRCh38). 1.1: Feature scanning and invalid region masking of a specific reference genome: The database construction module reads the input reference genome FASTA format sequence file. The FASTA format is a text format used to store nucleic acid or protein sequences. In this invention, the FASTA format is used to store the reference genome sequence. The file includes a description line for identifying chromosome or sequence names and the corresponding nucleotide sequence. Nucleotides are usually represented by A, T, C, and G, and N can be used to represent unknown or undetermined bases. The program script traverses the base sequence of the reference chromosome (taking human chromosome 21 as an example), identifies and filters out effective nucleotide intervals that are longer than 300 bp and do not contain any undetermined bases (i.e., invalid regions are removed, and the repaired region does not contain any invalid regions). Only the filtered effective intervals are used as candidate regions for subsequent spatial physical feature segmentation. 1.2: Non-overlapping grid segmentation along chromosome coordinate axes: The gridding module receives the valid intervals selected in the previous steps and, starting from the first nucleotide coordinate of each interval, segments the candidate intervals into fragments of specific base lengths along the positive direction of the chromosome coordinate axes (5' to 3' direction) according to a fixed baseline length (stepwise segmentation, segmenting into fragments of specific base lengths). Each fragment serves as a grid unit (with L... bin = 2000 bp, or 2kb, for example L bin For each grid cell, where L represents length and bin represents a single grid cell, a non-overlapping, continuous physical grid is created. For each discrete grid cell (Genomic Bin), the system allocates a globally unique identifier in memory (Grid ID: chr*:N1-N2, where chr* represents the chromosome origin of the grid cell, N1 is the starting coordinate of the grid cell, and N2 is the ending coordinate of the grid cell). 1.3: Database Permanent Storage: The identifiers of each genomic grid cell generated in step 1.2 are used as unique identifiers, and a mapping relationship is established between the grid cell and the corresponding chromosome starting and ending coordinates. This mapping relationship is stored in BED format. The grid cells and their corresponding coordinate information are serialized, written, and persistently stored in a non-transitory computer-readable storage medium (such as a server solid-state drive), forming a "Genomic Identifiability Grid Feature Database." This database can be independently packaged and distributed, serving as the physical validity basis for all downstream sequencing variant quality control analysis processes.

[0036] See Figure 5 and Figure 1Embodiment 2 of this invention demonstrates in more detail how the system directly calls the genome identifiability grid feature database and measured o-UMR for data quality control in a real production environment. 2.1: Raw Data Input, Alignment, and Grid Unit Loading: Obtain the raw short-read whole-genome sequencing data of the target sample (obtain the raw sequencing data of a specific sample). The raw data is stored in FASTQ format. FASTQ format is a text data format used to store raw sequence data from high-throughput sequencing. It typically consists of sequence identifier information, nucleotide sequences, quality value identifier information, and quality values ​​corresponding to each base in the nucleotide sequence, used to simultaneously record the nucleotide sequences obtained from sequencing and their base quality information. In this invention, FASTQ format is used to store the raw sequencing data generated by the sequencer and serves as input data for subsequent sequence alignment. Using BWA-MEM alignment software, the sequencing data is aligned to the specific reference genome described in Example 1 to obtain the alignment results between the sequencing sequence and the reference genome. These alignment results are then stored as a BAM format file. The BAM file is a binary data format used to store the alignment results between high-throughput sequencing sequences and the reference genome. The BAM format can record the alignment position, alignment direction, alignment quality value (MAPQ, Mapping Quality: a bioinformatics indicator that quantitatively describes the "confidence" or "accuracy" of a sequencing read being aligned to a specific position on the reference genome, based on the probability of alignment errors and calculated using the log-Phred scale), sequence information, and pairing relationships of paired-end sequencing fragments. In this invention, the BAM format is used to store the alignment results obtained after sequence alignment of FASTQ format raw sequencing data, providing a data foundation for subsequent statistical analysis of alignment information such as sequencing depth, unique aligned reads, multiple mapped reads, and MAPQ. Simultaneously, a pre-constructed genome identifiability grid feature database is loaded from the storage medium described in Example 1, and a correspondence is established between the genomic regions corresponding to the sequencing data and the grid units in the genome identifiability grid feature database based on the reference genome coordinate information of the target sample. 2.2: Variation Detection: The BAM format alignment file obtained in step 2.1 is used to perform variation detection using the variation detection software DeepVariant to obtain the genome variation detection results of the target sample. The variation detection results are stored in gVCF (genomic Variant Call Format), which is a text data format used to store genome variation detection results and genotype information of non-variant genomic regions, and is a data format used to record genotype information within the genome.In addition to recording variant sites such as SNV and INDEL and their genotypes, quality values, etc., gVCF can also record the genotypes and related quality information of non-variable regions through the reference interval, thereby realizing the recording of genotype information of continuous regions of the genome. In this invention, the gVCF format is used to save the genotypes and quality information such as GQ (Genotype Quality) of the target sample throughout the whole genome, providing a data basis for subsequent statistical analysis of the GQ values ​​of all grid units according to the genome grid units. 2.3: Statistical alignment information: The processor independently performs observational alignment uniqueness statistics for each grid unit, and performs concurrent fragment counting with two sets of parameter branches: (1) Statistically count the number of empirically unique mapped short fragments (R) that fall into the current grid unit, satisfy the correct pairwise pairing, and have an alignment quality score MAPQ > 30 (MAPQ, MappingQuality: a bioinformatics indicator based on the probability of alignment errors and calculated by the logarithmic Phred scale, used to quantitatively describe the "confidence" or "accuracy" of a sequencing read being aligned to a specific position on the reference genome). unique (1) The number of unique alignment reads in the grid cell; (2) Calculate the average sequencing depth of each grid cell: Call the coverage statistics program (e.g., bedtools coverage), use the genome grid cell as the statistics window, and use the alignment result file of the target sample as the coverage object to calculate the average sequencing coverage depth of all base sites in each grid cell. In this embodiment, the coverage calculation can be completed using, for example, the following command: bedtools coverage \;-a GenomicBin.bed \;-b Sample.bam \;-mean; Its mathematical expression is:

[0037]

[0038] Among them: MeanDepth i The average sequencing depth of the i-th grid cell; Depth j L represents the coverage depth of the j-th base site within a grid cell; bin The grid cell length is used. The resulting average sequencing depth is used as the coverage parameter for subsequent reliability risk mapping. Formula explanation and derivation: Calculate the average sequencing depth of this region, which is the average sequencing depth of all bases in that region. Enable multiple mapping fragment statistics to count the total number of short fragments (R0) allocated to this grid cell range. total The total number of reads for this grid cell is calculated as the number of uniquely mapped segments and all multi-mapped segments. 2.4: Calculation of the observation uniqueness ratio o-UMR: The system calculates the observation uniqueness ratio of this sample on the short read length platform using the following formula:

[0039]

[0040] Where: R unique R represents the number of uniquely matched reads within that grid cell. total , where is the total number of reads in the grid cell. If there is no data coverage in the grid, the value is 0. Formula derivation and explanation: When detecting simple genomic variations, the variation sites are often single-base (SNV) or multi-base (Indel) changes, but their length is much smaller than the 150bp of sequencing technology, making it difficult to determine whether this part of the detection results will be affected by multiple alignments. The sequence coordinates corresponding to a gene are often discontinuous on the genome, and some sequences may even be shared by multiple genes, making it difficult to quantify the multiple alignment situation of a gene. To address the above problems, the reference genome region can be divided into continuous, known specific length reads (e.g., 2000bp) manually or through parameter control to quantify the unique alignment situation on this sequence segment, thereby characterizing the identifiability of this sequence segment. Based on the annotation of each read segment in the alignment results by the alignment software, the number of uniquely aligned reads and the total number of reads within each 2000bp range are counted, and the ratio of the two is used as the observed unique alignment rate. 2.5: Weighted Calculation of Genotype Quality Scores for Grid Fragments: The system parses the variant detection file (gVCF format) output for this biological sample in step 2.2. Since the GQ (Genotype Quality Scores) representing genotype quality confidence in the gVCF format mixes single-base mutation sites with consecutive unmutated reference record intervals, directly performing an arithmetic average would systematically dilute the true quality scores of mutation sites in large reference blocks. This embodiment introduces a length-weighted averaging algorithm to calculate the overall genotype quality score of the 2kb grid.

[0041]

[0042] in:

[0043] bin represents a single grid cell; NGS represents next-generation sequencing technology; GQ(NGS, bin) represents the overall quality score of a single grid cell calculated based on the sequencing results from next-generation sequencing technology. i L is the GQ score of the i-th contiguous region in this grid cell; overlap,i L represents the length of the overlap between the i-th block within the grid cell and the grid cell, where overlap indicates the area where the block of the GQ score record overlaps with the grid cell; binThis corresponds to the length of the grid cell. The formula explanation and derivation are as follows: Because gVCF files record scores in "blocks" of different lengths, the quality score of a grid cell is calculated as follows to evaluate its quality score: multiply the quality score of each block within the grid cell by its overlap length within that cell, sum these results, and finally divide by the total length of the grid cell. This ratio is used as the final quality score of the grid cell. 2.6: See [link / reference] Figure 1 The risk mapping and quality control decision-making process of this invention is as follows: After obtaining the observed unique alignment rate (o-UMR), mean sequencing depth, and genotype quality score (GQ) for each genomic grid unit, the system further constructs a risk map to characterize the reliability of variant detection. Specifically, the system treats each genomic grid unit as an independent observation object and establishes an associated data structure composed of observed unique alignment rate, mean sequencing depth, and variant quality information: Ri = (o-UMR) / (mean depth / mean depth). i MeanDepth i GQ i Where: o-UMRi represents the unique alignment rate of the observation in the i-th grid cell; MeanDepth i GQ represents the average sequencing depth of the i-th grid unit; i This represents the genotype quality score of the i-th grid cell. To construct a risk map characterizing the reliability of variant detection, the observed unique alignment rate (o-UMR) is first plotted as the x-axis (a linear coordinate axis, ranging from [0-1.0]), and the logarithm of the average sequencing depth is plotted as log... 10 (MeanDepth) is the ordinate (logarithmically transformed coordinate axis, the value range is adaptively set according to the data distribution, such as

[10] ). 1 10 3 Each grid cell is mapped to a discrete circular data point in two-dimensional space, the coordinates of which are determined by the observational unique alignment rate (o-UMR) of that grid cell and the logarithm of the average sequencing depth of that grid cell. 10(MeanDepth) Decision. To simultaneously demonstrate the reliability of genotype determination, the genotype quality score (GQ) of each grid unit is mapped to the internal fill color or grayscale of the data point, establishing a continuous gradient from white or the first color (corresponding to low GQ) to black or the second color (corresponding to high GQ); all data points are uniformly outlined with black lines (width set to 0.3pt); the radius of the data points is uniformly set to 1.2 px to 1.5 px (preferably 1.5 px); the fill transparency is set to 0.6-0.9 (preferably 0.85); based on the preset horizontal axis boundary lines o-UMR = 0.25 and o-UMR = 0.75, black dashed lines are drawn, vertically dividing the entire map into Q1 high identifiability region, Q2 medium identifiability region, and Q3 low identifiability region, thereby forming a whole-genome reliability risk distribution map and completing the level classification of each grid unit.

[0044] (1) When o-UMR ≥ 0.75, it is determined to be a highly identifiable region;

[0045] (2) When 0.25 ≤ o-UMR < 0.75, it is judged as a medium identifiable region;

[0046] (3) When o-UMR < 0.25, it is determined to be a low identifiability region.

[0047] For variant sites in highly identifiable regions, the system retains their original quality assessment results and outputs them to the final variant result file. For variant sites in moderately identifiable regions, the system adds risk warning markers and retains them for subsequent manual review, auxiliary verification, or other purposes. For variant sites in low-identifiable regions, the system performs automated risk control processing, including but not limited to quality downgrading, risk marking, result filtering, or inclusion in the targeted rescue recommendation list. Furthermore, the system statistically analyzes the genomic coordinate intervals corresponding to all low-identifiable regions, automatically generates a standardized targeted rescue list, and outputs the list as a BED format file to guide subsequent long-read sequencing verification, targeted resequencing verification, or other independent verification processes. Through these steps, the system can uniformly map observational uniqueness alignment rate, sequencing coverage, and variant quality information to the same reliability assessment framework, achieving automated reliability assessment and quality control of whole-genome variant detection results.

[0048] See Figure 2Example 3 of the present invention describes the present invention based on the comparative analysis of highly identifiable regions and low-identifiable regions with natural depth differences. This example utilizes the inherent depth fluctuations of different genomic regions in whole-genome sequencing data to perform coverage-quality relationship analysis on grid cells of highly identifiable regions (Q1, o-UMR ≥ 0.75) and low-identifiable regions (Q3, o-UMR < 0.25) without artificial downsampling. This verifies the ability of observed unique alignment rate to characterize the reliability of variant detection and to verify the rationality and effectiveness of the quality control strategy of the present invention. 3.1: Depth Measurement and Genotype Quality Extraction of Genomic Grid Cells: The system obtains the raw alignment file (BAM format) of the whole-genome sequencing of the target sample on a short-read platform. Due to factors such as GC bias, local complex sequence structures, and random sampling effects during the sequencing process, different genomic regions will naturally form coverage differences, resulting in a continuous depth distribution from low-coverage regions to high-coverage regions. The system traverses all genomic grid units and extracts the following parameters: calculates the natural average sequencing depth (MeanDepth) of each grid unit; extracts the variant quality information of the grid unit; and calculates the genotype quality score (GQ) of each grid unit using the length-weighted average algorithm described in the previous embodiment. 3.2: Grouping of High-Identifiability and Low-Identifiability Regions: Based on the observational unique alignment rate obtained in Example 2, the system classifies all grid units according to a preset threshold. Specifically: when o-UMR ≥ 0.75, it is assigned to the high-identifiability region set Q1; when o-UMR < 0.25, it is assigned to the low-identifiability region set Q3. Subsequently, all grid units and their parameter information corresponding to sets Q1 and Q3 are extracted respectively. 3.3: Coverage-Quality Association Analysis: The system constructs a coverage-quality association matrix based on the grid units in sets Q1 and Q3. In this embodiment, the system uses average sequencing depth and genotype quality score as two independent analysis dimensions, mapping each grid unit to discrete observation points and constructing a corresponding scatter distribution. Subsequently, the Locally Weighted Regression (LOESS) algorithm was used to fit the scatter plots in the two regions respectively, obtaining the response curves of genotype quality scores as a function of sequencing depth. 3.4: Result analysis and quality control strategy verification: such as... Figure 2As shown, the system outputs and compares the coverage-quality response curves of regions Q1 and Q3. The analysis results show that: (1) In regions with high identifiability, the genotype quality score shows a continuous upward trend as the average sequencing depth increases, and gradually stabilizes after reaching a certain coverage, forming a high-quality platform region; (2) In regions with low identifiability, even if some grid units have a high sequencing depth, their genotype quality score remains at a low level or the improvement is limited; (3) The above phenomena indicate that in complex repetitive sequence regions, simply increasing the sequencing coverage cannot guarantee a synchronous improvement in variant detection quality, and regional identifiability has a significant impact on variant detection results; (4) The observed unique alignment rate can effectively distinguish genomic regions with different reliability levels and can be used as an important evaluation indicator for variant detection quality control. Therefore, the technical solution of this invention, which classifies genomic regions based on the observed unique alignment rate and implements variant result filtering, risk labeling, and targeted rescue analysis accordingly, has sufficient rationality and effectiveness.

[0049] See Figure 3 , Figure 4This is a feasible embodiment 4 for the long-read sequencing rescue and reconstruction of low-identifiability regions in Embodiment 2 of the present invention. P < 2e-16 indicates a significant difference. For the genomic grid cells identified as low-identifiability regions (o-UMR below a preset threshold) in Embodiment 2, the system can further utilize long-read sequencing data to perform targeted validation and quality reconstruction to recover variation information that is difficult to accurately resolve under short-read sequencing conditions. 4.1: Generation of a low-identifiability region targeted rescue list: The system reads the quality control results generated in Embodiment 2 and extracts the coordinate information of all grid cells marked as low-identifiability regions. For grid cells containing, for example, clinically relevant genes, drug metabolism genes, genetic disease-related loci, or other target regions, the system automatically summarizes the corresponding chromosome coordinate intervals and generates a standardized BED format targeted rescue recommendation list. The targeted rescue recommendation list can be used as an input file for subsequent long-read sequencing validation or targeted resequencing analysis. 4.2: Acquisition of long-read sequencing data: Acquire the long-read sequencing data corresponding to the target sample. The system performs alignment and variant detection analysis on long-read sequencing data, generating corresponding variant result files, including gVCF files. 4.3: Low-identifiability region targeted reanalysis; The system reads the BED format targeted rescue recommendation list generated in step 4.1, extracting alignment records and variant records covering the corresponding genomic coordinate regions from the long-read sequencing results. Because long-read sequencing reads can cross repetitive regions, paralogous regions, or complex structural regions that are difficult to uniquely locate in short-read sequencing, more complete local sequence information can be obtained. The system uses the extracted long-read variant results as the data basis for subsequent quality reconstruction. 4.4: Single-base resolution quality reconstruction; The system calls the single-base resolution quality reconstruction algorithm to map the quality scores in the long-read variant files to a single-base precision matrix, so that each nucleotide position corresponds to an independent quality score. In this embodiment, at nucleotide position j where sequence blocks overlap, a single-base level mathematical average operation is performed to obtain the fused single-base quality score GQ. base,j The overall long read variation quality score (GQ) of the reconstructed grid cell is calculated using the following formula:

[0050]

[0051] Among them: GQ base,j Let be the quality fraction of each block at the j-th base site, which is the average GQ score of all records at that position, where base represents the base; L bin This represents the length of the grid cell; bin indicates a single grid cell; TGS represents third-generation sequencing technology; This describes the overall quality score of a single grid cell calculated based on sequencing results from third-generation sequencing technology. Formula derivation and explanation: Because complex block overlap often exists in the gVCF result files of third-generation data analysis, direct summation would lead to an inflated quality score. Therefore, we first restore the quality score of each base before calculating the quality score of the grid cell, specifically as follows: For the j-th base site within the grid cell, extract the quality scores of all blocks covering that site and calculate the arithmetic mean to obtain the fusion quality score of that site. Then, calculate the arithmetic mean of the quality scores of all bases within the grid cell to obtain the overall quality score of the grid cell. 4.5: Comparative Analysis of Second- and Third-Generation Sequencing Results for Different Identifiable Regions: To quantitatively evaluate the quality reconstruction capability of long-read sequencing for variant detection results in different identifiable regions, after completing the risk map construction described in Example 2, the system extracted the sets of genomic grid units corresponding to high-identifiable regions (Q1, o-UMR≥0.75), medium-identifiable regions (Q2, 0.25≤o-UMR<0.75), and low-identifiable regions (Q3, o-UMR<0.25), respectively. For each grid unit, the following were statistically analyzed: (1) the genotype quality score GQShort obtained from short-read sequencing data; (2) the reconstructed genotype quality score GQLong obtained through the long-read sequencing quality reconstruction process described in Example 2. The system used grid units as statistical objects to group and compare GQShort and GQLong in regions Q1, Q2, and Q3, and used non-parametric statistical tests to evaluate the significance of the differences between the two groups of quality score distributions. See attached... Figure 3As shown, the quality improvement effect of long-read sequencing quality reconstruction was observed in different identifiable regions, and its specific manifestations are as follows: (1) High identifiable region (Q1): Due to the high observation unique alignment rate in this region, short-read sequencing has been able to obtain relatively stable variant detection results, and its genotype quality score is generally at a high quality plateau. After long-read sequencing reconstruction, the quality score is further improved, but the improvement is relatively limited, indicating that this region itself already has a good identifiable foundation. (2) Medium identifiable region (Q2). As the observation unique alignment rate decreases, the short-read sequencing results begin to fluctuate significantly, and some grid units show a decline in quality. After long-read sequencing reconstruction, the genotype quality score of most grid units is improved, and its quality distribution clusters towards the high-quality interval, indicating that long-read sequencing can effectively alleviate the measurement uncertainty caused by local sequence homology. (3) Low identifiable region (Q3): This region corresponds to highly repetitive sequences, highly paralogous gene clusters, or other complex genomic structures. Under short-read sequencing conditions, a large number of grid units exhibited significant quality collapse, with their genotype quality scores concentrated in low-quality regions. After long-read sequencing and single-base resolution quality reconstruction, the overall genotype quality score in this region significantly improved, and its quality distribution clearly shifted towards higher-quality regions, approaching the quality level corresponding to high-identifiability regions. 4.6: Reconstruction Result Output: The system correlates and compares the reconstructed variant quality score with the short-read sequencing results and generates corresponding reconstruction result files. For variant sites that are sufficiently supported after long-read sequencing verification, the system can use the reconstruction results to correct, supplement, or replace the short-read sequencing analysis results. Through the above steps, the system achieves supplementary verification and quality reconstruction of variant detection results in low-identifiability regions, thereby improving the reliability of variant detection in complex genomic regions and providing a data foundation for the subsequent construction of high-confidence variant sets. After obtaining the observed unique alignment rate (o-UMR), mean sequencing depth (MeanDepth), and genotype quality score (GQ) of each genomic grid unit after reconstruction, the system further constructs a risk map to characterize the reliability of variant detection, as shown in the appendix. Figure 4 As shown.

[0052] To better understand the various embodiments of the present invention, some original data samples from the actual working scenario of the present invention are provided in Table 1.

[0053] Table 1: Data Examples of Models Based on Sequence Identifiability

[0054] chr21 5030000 5032000 325 44 21.44 21.60 chr21:5030000-5032000 chr21 5032000 5034000 278 90 18.25 26.17 chr21:5032000-5034000 chr21 5034000 5036000 316 87 20.32 25.30 chr21:5034000-5036000 chr21 5036000 5038000 206 15 13.28 14.38 chr21:5036000-5038000 chr21 5038000 5040000 275 94 18.92 26.79 chr21:5038000-5040000 chr21 5048000 5050000 264 38 18.41 13.60 chr21:5048000-5050000 chr21 5050000 5052000 251 75 16.11 20.14 chr21:5050000-5052000 chr21 5058000 5060000 320 31 21.24 20.70 chr21:5058000-5060000 chr21 5062000 5064000 403 230 27.36 34.26 chr21:5062000-5064000

[0055] The present invention also provides a quality control device for genome variant detection based on sequence identifiability, comprising: a database management unit for storing and loading a genome identifiability grid feature database, the database containing continuous and non-overlapping grid cells segmented along chromosome coordinate axes at a preset baseline length and their corresponding physical coordinate identifiers; an observation unique alignment rate calculation unit for calculating the observation unique alignment rate of each grid cell in the target sample sequencing data based on the grid cells; a sequencing depth statistics unit for calculating the average sequencing depth of each grid cell; a quality parameter extraction unit for extracting variant quality information of each grid cell in the variant retrieval results; a risk map generation unit for constructing a whole-genome variant detection reliability risk map based on the observation unique alignment rate, the average sequencing depth, and the variant quality information; and a quality control unit for performing automatic labeling, downgrading, filtering, or rescue analysis on the variant detection results based on the risk map. The target rescue list generation unit is used to summarize the genomic coordinate ranges of low-identifiability regions and generate a standardized target rescue list; the long read data interface unit is used to acquire long read sequencing variant data; the quality reconstruction unit is used to map the long read variant quality score to the single base precision matrix and calculate the reconstruction quality score of the grid unit; the result fusion unit is used to correct, supplement or replace the short read variant detection results based on the reconstruction quality score.

[0056] This invention also provides a method for constructing a grid feature database for building a benchmark for genome sequencing analysis, executed by a computer device, characterized by comprising the following steps: obtaining a specific reference genome sequence; scanning the specific reference genome sequence; identifying and extracting effective sequencing intervals with consecutive non-N bases and a length greater than a preset physical threshold; and removing unknown regions; dividing the effective sequencing intervals into continuous and non-overlapping grid units along the chromosome coordinate axes according to a preset benchmark length; assigning a unique physical coordinate identifier to each grid unit and storing the physical coordinate identifier in association with chromosome coordinate information; and constructing a genome identifiability grid feature database, wherein the database is used for subsequent gridded statistics of sequencing reads, calculation of observational unique alignment rates, and reliability assessment of variant detection.

[0057] The present invention may also provide a computer-readable storage medium storing a genome identifiable grid feature database, the database containing multiple structured data entries, each data entry being associated with at least: the physical coordinate identifier of the grid cell and chromosome coordinate information; when the genome identifiable grid feature database is accessed by an external sequencing data analysis system, it provides a unified genome coordinate index.

[0058] The methods and processes for constructing various maps and related quality control decisions, evaluations, judgments, detections, and comparative analyses of the present invention can also be constructed as one or more other embodiments that are automatically implemented by a program. These embodiments can be presented and implemented using visualization methods, and are realized based on the aforementioned methods and processes of the present invention, including drawing, coloring, and judging related elements.

[0059] The above description is merely a preferred embodiment of the present invention for reliable detection and evaluation of genomic variations based on sequence identifiability, and does not constitute any limitation on the structure of the present invention. These embodiments can also be split or combined to form new embodiments depending on specific circumstances. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention fall within the scope of the present invention. As mentioned above, for a better understanding of the present invention, the following references can be consulted, the entire contents of which are incorporated herein by reference.

[0060] 1. "Bioinformatics", Chen, M., Science Press, 4th ed., (2022).

[0061] 2. "Genomic darkmatter: the reliability of short read mapping illustrated by the genomemappability score", Lee, H. and Schatz, MC, in Bioinformatics, vol.28, pp.2097-2105, (2012)

[0062] 3. "Accurate circular consensus long-read sequencing improves variant detection and assembly of a human genome", Wenger, AM et al., Nature Biotechnology, vol.37, pp.1155-1162, (2019).

[0063] 4. "The sequence of the human genome", Venter, JC et al., in Science, vol.291, pp.1304-1351, (2001).

[0064] 5. "The complete sequence of a human genome", Nurk, S. et al., Science, vol.376, pp.44-53, (2022).

[0065] 6. "Telomere-to-telomere assembly of a complete human X chromosome", Miga, KH et al., Nature, vol.585, pp.79-84, (2020)

[0066] 7. "Genetic variation, comparative genomics, and the diagnosis of disease", Eichler, EE, in New England Journal of Medicine, vol.381, pp.64-74, (2019)

[0067] 8. "Toward better understanding of artifacts invariant calling from high-coverage samples", Li, H., in Bioinformatics, vol.30, pp.2843-2851, (2014)

[0068] 9. "The structure, function and evolution of a complete human chromosome 8", Logsdon, GA et al., in Nature, vol.593, pp.101-107,(2021)).

Claims

1. A method for quality control detection of genomic variations, characterized in that, include: A reference genome sequence is obtained, and the reference genome sequence is divided into continuous and non-overlapping grid units along the chromosome coordinate axis according to a preset baseline length. A unique physical coordinate identifier is assigned to each grid unit, and the physical coordinate identifier is stored in association with chromosome coordinate information to construct a genome identifiability grid feature database. The sequencing data alignment file of the target sample is obtained. Using the grid cell as the statistical unit, the observation unique alignment rate of each grid cell is calculated. The observation unique alignment rate is the ratio of the number of unique mapped reads that fall within the grid cell and meet the preset alignment conditions to the total number of reads allocated to the grid cell. At the same time, the average sequencing depth corresponding to each genome grid cell is calculated. The initial gene mutation file for the target sample output by the upstream mutation detection tool is extracted, and the genotype quality score of each grid cell is calculated and fused using a length-weighted average algorithm as the mutation quality information of the grid cell. Based on the observed unique alignment rate, the average sequencing depth, and the variant quality information, a risk map is constructed to characterize the reliability of variant detection in each grid unit. Based on the risk map, quality control processing is performed on the mutation detection results.

2. The genomic variation quality control detection method according to claim 1, characterized in that, Before dividing the reference genome sequence into continuous and non-overlapping grid units along the chromosome coordinate axes according to a preset baseline length, the process further includes: The reference genome sequence is scanned to identify and extract valid sequencing intervals that are consecutive non-N bases and have a length greater than a preset physical threshold, while unknown regions containing N bases are removed; the grid unit is split only within the valid sequencing intervals.

3. The genomic variation quality control detection method according to claim 1, characterized in that, The observation uniqueness ratio o-UMR is calculated according to the following formula: ， R unique ,i represents the number of unique mapped reads within the i-th genome grid cell; R total ,i represents the total number of reads assigned to the i-th genomic grid unit.

4. The genomic variation quality control detection method according to claim 1, characterized in that: The observational unique alignment rate (o-UMR) and average sequencing depth were used as two independent evaluation dimensions, and the genotype quality score of the corresponding grid unit was represented by a color gradient, thus forming a whole-genome reliability risk distribution map. Each grid cell corresponds to a discrete observation point in the map; based on a preset observation uniqueness comparison rate threshold, each grid cell is divided into: high identifiability region; medium identifiability region; and low identifiability region. When o-UMR ≥ 0.75, it is defined as a highly identifiable region; When 0.25 ≦ o-UMR < 0.75, it is defined as a region of moderate identifiability; When o-UMR < 0.25, it is defined as a low identifiability region.

5. The genomic variation quality control detection method according to claim 4, characterized in that: Automatically mark risk warnings for each point; Summarize the genomic coordinate intervals corresponding to low-identifiability regions; A targeted rescue recommendation list containing the genomic coordinate range information is generated to guide subsequent validation sequencing, resequencing analysis, or cross-sequencing platform validation analysis.

6. The genomic variation quality control detection method according to claim 4, characterized in that: Genome grid units with low identifiability are used to perform targeted validation and quality reconstruction using long-read sequencing data to execute a long-read rescue process, including the following steps: Generating a list of targeted rescue operations for low-identifiability areas; Based on the generated quality control results, extract the coordinate information of all grid cells marked as low identifiability areas; For grid cells containing clinically relevant genes, drug metabolism genes, genetic disease-related loci, or other target regions, the system automatically summarizes the corresponding chromosome coordinate intervals and generates a standardized BED format targeted rescue recommendation list. The targeted rescue recommendation list can be used as an input file for subsequent long-read sequencing validation or targeted resequencing analysis; Long-read sequencing data acquisition; Obtain long-read sequencing data corresponding to the target sample; Perform alignment and variant detection analysis on long-read sequencing data to generate corresponding variant result files, including gVCF files; Targeted reanalysis of low-identifiability regions; The system reads the generated BED format targeted rescue recommendation list and extracts alignment records and variant records covering the corresponding genomic coordinate regions from the long-read sequencing results; Single-base resolution quality reconstruction; The single-base resolution quality reconstruction algorithm is called to map the quality scores of long-read variant files to the single-base precision matrix, so that each nucleotide position corresponds to an independent quality score. At nucleotide position j where sequence block overlap occurs, perform a single-base-level mathematical averaging operation to obtain the fused single-base quality score GQ. base,j ; The overall long read variation quality score (GQ) of the reconstructed grid cell is calculated using the following formula: ； Comparative analysis of the quality before and after rescue of different identifiable areas: Genome grid unit sets corresponding to high identifiability region Q1, medium identifiability region Q2, and low identifiability region Q3 were extracted respectively; For each grid cell, the following statistics are performed: (1) Genotype quality score (GQShort) obtained from short-read sequencing data; (2) The quality score (GQLong) of the reconstructed genotype obtained by the long-read sequencing quality reconstruction process; The system uses grid cells as the statistical object to group and compare GQShort and GQLong in high identifiability region Q1, medium identifiability region Q2 and low identifiability region Q3, and uses nonparametric statistical test methods to evaluate the difference between the quality score distributions of the two groups. Output the reconstruction results; The reconstructed variant quality score is correlated and compared with the short read sequencing results, and the corresponding reconstructed result file is generated. For variant sites that have been adequately supported by long-read sequencing, the reconstruction results are used to correct, supplement, or replace the short-read sequencing analysis results.

7. A genomic variation quality control detection system, characterized in that, include: A database management unit is used to store and load a genome identifiability grid feature database, which contains continuous and non-overlapping grid cells segmented along chromosome coordinate axes at a preset baseline length and their corresponding physical coordinate identifiers. The observation unique alignment rate calculation unit is used to calculate the observation unique alignment rate of each grid cell in the target sample sequencing data based on the grid cell as a statistical basis. Sequencing depth statistics unit, used to calculate the average sequencing depth of each grid cell; The quality parameter extraction unit is used to extract the mutation quality information of each grid cell in the mutation call result; The risk map generation unit is used to construct a reliability risk map for whole-genome variant detection based on the observed unique alignment rate, the average sequencing depth, and the variant quality information. The quality control unit is used to perform automatic labeling, downgrading, filtering, or rescue analysis on the mutation detection results based on the risk map.

8. The genomic variation quality control detection system according to claim 7, characterized in that, Also includes: The target rescue list generation unit is used to summarize the genomic coordinate ranges of low-identifiability regions and generate a standardized target rescue list; Long read data interface unit, used to acquire long read sequencing variant data; The quality reconstruction unit is used to map the long read variation quality score to the single base accuracy matrix and calculate the reconstruction quality score of the grid cell. The results fusion unit is used to correct, supplement, or replace the short read variation detection results based on the reconstruction quality score.

Images