A method for identifying genomic structural variations based on long-read sequencing

Abstract

The application provides a genome structural variation identification method based on long read sequencing, and belongs to the technical field of bioinformatics. The method comprises the following steps: obtaining long read sequencing data and performing pretreatment, aligning high-quality reads to a reference genome to screen a candidate variation region, and further constructing a contig sequence based on local assembly; combining multi-dimensional features such as sequence similarity, breakpoint position, coverage depth and repeat structure to establish an alignment model, and calculating a breakpoint support number to realize accurate positioning of the breakpoint; and finally identifying structural variation types such as insertion, deletion, inversion and repeat amplification. The application can effectively improve the detection capability of complex structural variations and repeat regions, and has high accuracy and practicability.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics technology, and in particular relates to a method for identifying genomic structural variations based on long-read sequencing. Background Technology

[0002] With the development of high-throughput sequencing technology, genomic variation detection has become one of the core technologies in hereditary disease research, tumor genomics, and precision medicine. Genomic variations typically include single nucleotide variants (SNVs), small fragment insertions and deletions (Indels), and structural variations (SVs). Among these, structural variations mainly include large fragment insertions, deletions, inversions, duplications, and translocations, which play an important role in gene expression regulation, chromosome structural stability, and disease development. Numerous studies have shown that structural variations are closely related to a variety of complex diseases, but due to their structural complexity and wide range, their detection is significantly more difficult than that of SNVs and small fragment variations.

[0003] Current methods for detecting genomic variations primarily rely on next-generation sequencing (NGS) technology. This technology identifies potential variant sites by acquiring large amounts of short-read sequence data and aligning them to a reference genome. NGS technology is relatively mature in detecting SNVs and small insertions / deletions, offering high accuracy and throughput. However, its typically short read lengths (generally 100-300 bp) present significant limitations when dealing with complex structural variations. For example, in repetitive regions, low-complexity regions, and "dark areas" of the genome, short-read data are difficult to align uniquely, easily leading to alignment errors or information loss, thus reducing the sensitivity and accuracy of structural variation detection. Furthermore, short-read-based analysis methods often rely on single or limited features such as split-read alignment, paired-end alignment, and changes in coverage depth for variation inference, which are insufficient for analyzing complex structural variations (such as nested repeats and long fragment inversions).

[0004] In recent years, with the development of third-generation sequencing technology, long-read sequencing (such as single-molecule real-time sequencing and nanopore sequencing) has been gradually applied to genome research. This type of technology can obtain reads in lengths of thousands of base pairs or even longer, theoretically capable of traversing complex repetitive regions and directly covering structural variation breakpoints, providing a new technical pathway for structural variation detection. However, long-read sequencing data typically have a high raw error rate, and its data analysis methods are still imperfect. How to effectively utilize long-read data to achieve high-precision structural variation identification remains a pressing technical problem to be solved in this field.

[0005] Some existing methods attempt to detect structural variations using long-read data, but most still follow the traditional alignment analysis approach, directly aligning reads to a reference genome and identifying variations based on alignment anomalies. These methods fail to fully leverage the advantages of long reads in traversing complex structures, particularly in regions of repeated amplification and structural complexity, where issues such as inaccurate breakpoint localization and high false positive rates persist. Furthermore, while some methods incorporate local assembly strategies, they often lack systematic modeling and multi-feature fusion analysis of the assembly results, leading to reliance on single evidence and insufficient robustness in structural variation determination.

[0006] Therefore, how to construct a structural variation identification method that can fully utilize the advantages of long-read sequencing data, take into account data error correction, local structure reconstruction, and multi-dimensional feature comprehensive analysis, so as to achieve high-precision detection of complex structural variations and accurate location of breakpoints, has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] The first objective of this invention is to propose a method for identifying genomic structural variations based on long-read sequencing, comprising the following steps: S1 acquires long-read sequencing data of the target sample, wherein the long-read sequencing data is raw reads data with a single read length greater than 1kb; S2 preprocesses the long-read sequencing data, including noise reduction, quality filtering, and error correction, to obtain a high-quality set of reads; S3 aligns the high-quality reads to the reference genome and screens out candidate variant regions with alignment anomalies, including soft splicing, break alignment, or cross-region alignment. S4 extracts the Reads covering the candidate mutation region and performs local assembly to construct the contig of the candidate mutation region. S5 establishes a multi-dimensional feature alignment model based on the alignment relationship between the contiguous group and the reference genome, wherein the multi-dimensional features include at least one of the following: Sequence similarity features; Compare the breakpoint location features; Read length coverage depth features; Structural features of repeating sequences; S6 calculates the number of breakpoint support points in the candidate mutation region based on the multi-dimensional feature alignment model, and locates the breakpoints. S7 identifies the types of structural variations in the target genome and their corresponding locations based on the breakpoint location and its number of supports. The types of structural variations include at least insertions, deletions, inversions, and duplications.

[0008] Preferably, the error correction in step S2 includes the following steps: Construct an overlap graph based on the overlap relationships between Reads; For each Read, identify its set of candidate overlapping Reads, wherein the candidate overlapping Reads satisfy the condition that the overlap length is greater than a preset threshold and the sequence similarity is higher than a preset threshold; A consensus sequence is generated based on the candidate overlapping Reads set; The consensus sequence is used to replace the incorrect bases in the original Reads to obtain the corrected Reads; The preset threshold includes: The overlap length threshold is above 500bp; The sequence similarity threshold is above 80%.

[0009] 3. The method for identifying genomic structural variations according to claim 1, characterized in that the screening of candidate variation regions in step S3 includes the following steps: 1) Analyze the alignment results of all Reads and extract the following abnormal alignment features: Length of soft-clipped fragments; Split-read location and its cross-region distance; Coverage depth distribution of Reads on the reference genome; 2) Based on the aforementioned anomaly comparison features, candidate mutation regions are selected according to the following rules: When there is at least one read with a soft-splitting fragment length greater than a preset length threshold, and the soft-splitting sites form local clusters in the genome, the region is marked as a candidate region. Alternatively, if there are at least two split alignment reads whose alignment positions are located in non-contiguous regions of the reference genome and the distance exceeds a preset distance threshold, the region is marked as a candidate region. Alternatively, if the deviation of the target region's coverage depth from the average coverage depth of the whole genome exceeds a preset threshold, the region is marked as a candidate region. 3) Merge regions that satisfy any of the above rules into candidate variant regions.

[0010] Preferably, the partial assembly in step S3 includes the following steps: 1) Extract Reads that meet the following conditions from the candidate mutation regions as seed Reads: Cover the candidate mutation region; It may contain soft-sheared fragments or split alignment information; 2) Based on the seed Reads, construct a local overlap map according to the overlap relationship between the Reads; 3) Perform path search and extension operations in the overlapping graph to generate at least one candidate contig sequence; 4) Perform consistency correction on the candidate contig sequences to obtain the target contig sequence; 5) Use the target contig sequence for subsequent breakpoint identification.

[0011] Preferably, the multi-dimensional feature alignment model in step S5 includes the following steps: 1) Align the target contig sequence obtained in step 4) with the reference genome to obtain the initial alignment result; 2) Based on the alignment results, extract the following feature parameters: Sequence similarity features are used to characterize the degree of matching between the contig sequence and the reference sequence; Breakpoint location features are used to characterize the coordinates and distribution of discontinuous positions in the comparison. Coverage depth feature is used to characterize the number and distribution of Reads supporting the contig; Repeating structural features are used to characterize the length of repeating sequences and the characteristics of repeating units in the target region; 3) Normalize the above feature parameters to obtain standardized feature vectors; 4) Calculate the comprehensive alignment score based on the standardized feature vectors; 5) Based on the comprehensive alignment score and breakpoint location characteristics, determine the breakpoint location of the candidate structural variation region.

[0012] The sequence similarity features are calculated based on the ratio of matching bases or alignment scores; The breakpoint location features are determined based on the gap, soft shear location, or splitting alignment location in the comparison. The coverage depth feature is calculated based on the number of Reads supporting the region or the coverage uniformity. The repeating structural features are calculated based on the changes in the length and copy number of the repeating unit.

[0013] Preferably, the calculation of the breakpoint support number in step S6 includes: 1) Based on the candidate breakpoint locations determined in step S5, extract all Reads and contig sequences that cross or are adjacent to the breakpoints; 2) Determine whether each Read or contig supports the breakpoint. The determination rules include: Reads are split and aligned at this position; Or Reads may contain soft shear structures that cross the break point; Or the contig sequence may have an alignment discontinuity at this position; 3) Count the number of Reads and contigs that satisfy the judgment rules, and use them as the support number of the breakpoint.

[0014] Preferably, the rules for determining the type of structural variation include: When an extra sequence is inserted, it is considered an insertion mutation. When there are consecutive deletions in the reference sequence, it is determined to be a deletion variant; When the sequence direction is reversed, it is determined to be an inversion mutation; When the number of repeating units increases, it is determined to be a repeat amplification mutation.

[0015] A second objective of this invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for identifying genomic structural variations.

[0016] The beneficial technical effects of the present invention are at least as follows: This invention, by introducing long-read sequencing data and combining it with a local assembly strategy, can effectively cross repetitive sequence regions and "dark areas" of the genome that are difficult to cover by traditional short-read sequencing, and significantly improve the detection capability and integrity of complex structural variations (especially large fragment insertions, inversions and duplication amplifications).

[0017] This invention constructs a multi-dimensional feature alignment model based on contigs, which integrates and analyzes various information such as sequence similarity, breakpoint location, coverage depth, and repeating structures. Compared with a single comparison strategy, this improves the accuracy and robustness of structural variation identification and reduces the probability of false detection and false negative detection.

[0018] This invention introduces a breakpoint support number mechanism and combines multi-source evidence from Reads and contigs for statistical analysis, thereby achieving reliable verification and precise localization of breakpoints and significantly improving the accuracy of structural variation breakpoint localization.

[0019] This invention adopts a technical approach of "local assembly + breakpoint-guided analysis", which avoids the high computational cost of whole genome assembly while ensuring computational efficiency, making the method highly practical and scalable in large-scale genome data analysis.

[0020] The overall method of this invention is clear and highly feasible, and can be widely applied in fields such as hereditary disease research, tumor genome analysis and precision medicine, providing a reliable technical means for the highly sensitive detection of complex structural variations. Attached Figure Description

[0021] Figure 1 This is a diagram illustrating breakpoint identification. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0024] Example 1: Basic Process for Structural Variation Identification Based on Long Read Sequencing This embodiment is used to verify the complete process of the method described in this invention and its ability to identify structural variations.

[0025] 1. Data Acquisition A human genome sample was selected as the target sample and sequenced using a long-read sequencing platform (such as PacBio or Oxford Nanopore) to obtain raw reads. The average length of the reads was approximately 10 kb, the shortest read was greater than 1 kb, and the total sequencing depth was approximately 30×.

[0026] 2. Data Preprocessing Preprocessing the raw Reads includes: 1) Quality Filtering: Removes Reads with an average quality value lower than Q10; 2) Denoising: Remove obviously abnormal low-complexity sequences; 3) Error correction: Construct an overlap graph based on the overlap relationships between Reads; Identify candidate overlapping reads with an overlap length greater than 500bp and a sequence similarity greater than 80%; Generate a consensus sequence based on candidate Reads; Replace the incorrect bases in the original Reads with the consensus sequence; The processed data yields a high-quality collection of Reads.

[0027] 3. Comparison and candidate region selection High-quality reads were aligned to the reference genome (GRCh38) using a long read alignment algorithm (minimap2).

[0028] Analyze the comparison results and extract abnormal features: Length of soft-clipped segments; Split-read alignment location and cross-region distance; Coverage depth distribution; The filtering rules are as follows: If the soft-clipped length is greater than 100bp and there are at least two clustered reads within a 1kb window, it is marked as a candidate region; Or there may be ≥2 split-reads whose alignment positions span a distance >1kb; Or the coverage depth deviates from the whole genome average by more than 50%; Regions that meet the criteria are merged to obtain a set of candidate variant regions.

[0029] 4. Partial assembly For each candidate mutation region: 1) Extract seed Reads, including: Reads covering this area; Or Reads that contain soft-clipped or split-read information; 2) Construct a local overlap map based on the overlap relationship of Reads; 3) Perform path extension in the overlap graph to generate candidate contig sequences; 4) Perform consistency correction on the contig to obtain the target contig; 5) Retain the information at both ends of the contig for breakpoint analysis.

[0030] 5. Multidimensional Feature Alignment Analysis The target contig was aligned to the reference genome, and the following features were extracted: Sequence similarity (ratio of matching bases); Breakpoint location (gap, soft-clipped, or split location); Coverage depth (number of supported Reads); Repeating structures (length of repeating units and number of copies); Each feature is normalized to construct a feature vector, and a comprehensive alignment score is calculated.

[0031] Candidate breakpoints are determined based on scoring and comparison of discontinuous locations.

[0032] 6. Calculation of the number of support points at the breakpoint For each candidate breakpoint: 1) Extract Reads and contigs that cross or are adjacent to the breakpoint; 2) Determine the supporting conditions: There is a split-read; There may be a soft-clipped structure across breakpoints; Or the contig alignment is discontinuous; 3) Count the number of supported Reads and contigs as the number of breakpoints supported.

[0033] When the number of supports is greater than or equal to 3, it is determined to be a valid breakpoint.

[0034] 7. Structural variation identification Based on the location of the breakpoint and the number of supports, determine the type of structural variation: An extra sequence exists → insertion; Reference sequence missing → missing; Reverse the sequence direction → invert; Increase in repeating units → Repeated amplification; The final output shows the type of structural variation and its precise location in the genome.

[0035] like Figure 1 As shown, the test data was analyzed using the above method, and various types of structural variations were successfully detected, including insertions, deletions, and duplications. The results indicate that this method can stably construct contig sequences within candidate variation regions and effectively identify breakpoints based on a multidimensional feature alignment model. Structural variations longer than 1 kb can be accurately detected, with breakpoint localization accuracy controlled within ±50 bp.

[0036] Table 1. Statistics of structural variation detection results in this embodiment. As shown in Table 1, in this embodiment, the structural variant identification method based on long-read sequencing detected multiple types of structural variants, including insertions, deletions, inversions, and duplications. All types of structural variants received support from multiple reads or contigs, and the number of breakpoints supported was generally above three, indicating that the identified structural variants have high reliability.

[0037] Further analysis shows that this method demonstrates good accuracy in locating breakpoints for different types of structural variants, with the breakpoint localization error controlled within ±50 bp. The localization accuracy is relatively higher for deletion-type variants. For complex structural variants such as duplication amplification, this method, through local assembly and multidimensional feature alignment analysis, can still obtain stable supporting evidence and achieve breakpoint localization.

[0038] Example 2: Complex Structural Variation (Repetition Amplification) Identification Scenario This embodiment is used to verify the recognition ability of the method of the present invention in complex structural variations, especially in regions of repetitive sequence amplification.

[0039] 1. Data Acquisition Genomic samples containing known duplication amplification variants were selected as target samples and sequenced using a long-read sequencing platform to obtain raw reads. The average length of the reads was approximately 12 kb, the shortest read was greater than 1 kb, and the total sequencing depth was approximately 35×.

[0040] The target region is a genomic region containing tandem repeat units, with each repeat unit being approximately 90 bp in length and having an increased copy number compared to the reference genome.

[0041] 2. Data Preprocessing The original Reads undergo quality filtering, noise reduction, and error correction, the specific process of which is the same as in Example 1 and will not be repeated here. A high-quality set of Reads is obtained after processing.

[0042] 3. Comparison and candidate region selection Align high-quality reads to the reference genome.

[0043] The following anomalous features were observed in the target repeating region: Multiple reads show obvious soft-clipped segments (length > 150bp) at the boundaries of repeated regions. There are split-reads that span overlapping regions, and their alignment positions are discontinuously distributed. The coverage depth within the region is significantly higher than the average level of the whole genome (approximately 80% higher); Based on the above characteristics, and in accordance with the preset screening rules, the repeated region was successfully marked as a candidate variant region.

[0044] 4. Partial assembly For the candidate region: 1) Extract Reads that cover the region and contain anomaly comparison features as seed Reads; 2) Construct a local overlap map based on the overlap relationship of Reads; 3) Perform path search in the overlap graph to generate multiple candidate contig sequences; 4) After performing consistency correction on the contig, the target contig is obtained; The contig spans both boundaries of the repeating region and preserves the continuous structural information of the repeating units.

[0045] 5. Multidimensional Feature Alignment Analysis Align the target contig to the reference genome and extract multidimensional features: Sequence similarity: High matching degree in non-repetitive regions, and local discontinuities appear in repetitive regions; Breakpoint location: There is a clear gap and comparison interruption at the boundary of the repeated region; Coverage depth: The number of Reads corresponding to repeated regions increases significantly; Repeating structure: The length of the repeating unit was stable, but the copy number increased; After normalizing the features, a comprehensive alignment score is calculated, and the candidate breakpoints are located at the boundaries of the repeated regions.

[0046] 6. Calculation of the number of support points at the breakpoint Regarding the above candidate breakpoints: Extract Reads and contigs that cross breakpoints; Identify supporting evidence, including: Split-read support; soft-clipped support; contig supports cross-breakpoint support; Statistics show that: Each breakpoint supports 5–9 reads; The number of contigs is 2–3; If the number of supported points meets the preset threshold, it is determined to be a valid breakpoint.

[0047] 7. Structural variation identification Based on the location of the breakpoint and the characteristics of the repeating structure, it was determined that there is a repeating amplification-type structural variation in this region.

[0048] Further analysis shows that: The repeating unit length is approximately 90 bp; The number of duplicate copies increased by approximately 2–3 times compared to the reference genome; The final output is the type of structural variation and its precise genomic location.

[0049] 8. Implementation Results The results of this embodiment show that, in repetitive sequences and highly complex regions, the method of the present invention can still construct contig sequences that span repetitive regions through local assembly, and effectively identify breakpoint locations by combining a multidimensional feature alignment model.

[0050] Compared to methods that rely solely on read alignment, this method avoids alignment ambiguity caused by repetitive sequences, exhibiting higher stability and accuracy in detecting repetitive amplification structural variations.

[0051] By statistically analyzing the number of breakpoints, the identification results can be effectively verified, improving the reliability of breakpoint determination. This verifies the applicability and technical advantages of the method of this invention in complex structural variation scenarios.

[0052] Example 3: Comparison and verification of this method with methods based solely on alignment. This embodiment is used to compare and verify the differences between the method of the present invention and the structural variation identification method based solely on Reads alignment in the detection of complex structural variations.

[0053] 1. Data Source Genomic sample data containing repeat amplification structural variations were selected as the test subjects. The sequencing data were long-read sequencing data, with an average read length of approximately 10kb, a minimum read length of more than 1kb, and a total sequencing depth of approximately 30×.

[0054] 2. Method settings (1) Method of the present invention The following steps were used to identify structural variations in the sequencing data: 1) Perform quality filtering, noise reduction, and error correction on the raw Reads data to obtain high-quality Reads; 2) Align high-quality reads to the reference genome, analyze the alignment results and extract abnormal alignment features, including soft-clipped fragments, split-read locations and coverage depth distribution, and screen candidate variant regions; 3) For candidate variant regions, extract reads that cover the region or contain abnormal alignment features, construct a local overlap map based on the overlap relationship between reads, and extend the path to generate contig sequences; 4) The contig sequence is compared with the reference genome to extract multidimensional features such as sequence similarity, breakpoint location, coverage depth and repetitive structure, and then normalized and comprehensively scored to determine the candidate breakpoint location. 5) Extract the Reads and contig sequences that cross or are adjacent to the breakpoint, count the number of Reads and contigs that support the breakpoint, and calculate the breakpoint support count; 6) Determine the type of structural variation and its location in the genome based on the location of the breakpoint and its number of supports.

[0055] (2) Comparison-based methods only The following steps are used to identify structural variations: 1) After quality filtering of sequencing reads, they are aligned to the reference genome; 2) Identify candidate breakpoints based on the split-read and coverage depth anomalies in the comparison results; 3) Determine the type of structural variation directly based on the location of the fracture point; This method does not include a local assembly step, nor does it perform contig-based multidimensional feature alignment analysis.

[0056] 3. Comparison Results Table 2 Comparison of detection performance between our method and methods based solely on alignment 4. Results Analysis In regions of repetitive amplified structural variations, methods based solely on read alignment struggle to accurately determine breakpoint locations due to multiple alignment issues caused by repetitive sequences. Furthermore, the consistency of alignment results between different reads is poor, leading to unstable structural variation identification results.

[0057] The method of this invention introduces a local assembly step to construct a contig sequence that spans repeating regions, thereby restoring the continuous sequence structure of the target region. At the same time, it combines multidimensional feature alignment analysis to comprehensively evaluate the breakpoint location, sequence similarity, coverage depth, and repeating structure, thereby improving the accuracy of breakpoint identification.

[0058] Furthermore, by statistically analyzing the number of support points for breakpoints, this method can utilize evidence provided by multiple Reads and contigs to verify breakpoints, effectively reducing the probability of misjudgment.

[0059] 5. Conclusion This embodiment demonstrates that, compared to structural variation identification methods based solely on Reads alignment, the method of this invention has higher detection capability and breakpoint localization accuracy in the detection of complex structural variations, especially in the detection of repeating amplification regions, and can effectively improve the stability and reliability of structural variation identification.

[0060] The above detailed description is a specific description of one of the feasible embodiments of the present invention. This embodiment is not intended to limit the patent scope of the present invention. All equivalent implementations or modifications that do not depart from the present invention should be included within the scope of the technical solution of the present invention.

Claims

1. A method for identifying genomic structural variations based on long-read sequencing, characterized in that, Includes the following steps: S1 acquires long-read sequencing data of the target sample, wherein the long-read sequencing data is raw reads data with a single read length greater than 1kb; S2 preprocesses the long-read sequencing data, including noise reduction, quality filtering, and error correction, to obtain a high-quality set of reads; S3 aligns the high-quality reads to the reference genome and screens out candidate variant regions with alignment anomalies, including soft splicing, break alignment, or cross-region alignment. S4 extracts the Reads covering the candidate mutation region and performs local assembly to construct the contig of the candidate mutation region. S5 establishes a multi-dimensional feature alignment model based on the alignment relationship between the contiguous group and the reference genome, wherein the multi-dimensional features include at least one of the following: Sequence similarity features; Compare the breakpoint location features; Read length coverage depth features; Structural features of repeating sequences; S6 calculates the number of breakpoint support points in the candidate mutation region based on the multi-dimensional feature alignment model, and locates the breakpoints. S7 identifies the types of structural variations in the target genome and their corresponding locations based on the breakpoint locations and their support numbers. The types of structural variations include at least insertions, deletions, inversions, and duplications.

2. The method for identifying genomic structural variations according to claim 1, characterized in that, The error correction in step S2 includes the following steps: Construct an overlap graph based on the overlap relationships between Reads; For each Read, identify its set of candidate overlapping Reads, wherein the candidate overlapping Reads satisfy the conditions that the overlap length is greater than a preset threshold and the sequence similarity is higher than a preset threshold; A consensus sequence is generated based on the candidate overlapping Reads set; The consensus sequence is used to replace the incorrect bases in the original Reads to obtain the corrected Reads; The preset threshold includes: The overlap length threshold is above 500bp; The sequence similarity threshold is above 80%.

3. The method for identifying genomic structural variations according to claim 1, characterized in that, Step S3, which involves screening candidate variant regions, includes the following steps: 1) Analyze the alignment results of all Reads and extract the following abnormal alignment features: Length of soft-clipped fragments; Split-read location and its cross-region distance; Coverage depth distribution of Reads on the reference genome; 2) Based on the aforementioned anomaly comparison features, candidate mutation regions are selected according to the following rules: When at least one read has a soft-splitting fragment length greater than a preset length threshold, and the soft-splitting sites form local clusters in the genome, the region is marked as a candidate region. Alternatively, if there are at least two split alignment reads whose alignment positions are located in non-contiguous regions of the reference genome and the distance exceeds a preset distance threshold, the region is marked as a candidate region. Alternatively, if the deviation of the target region's coverage depth from the average coverage depth of the whole genome exceeds a preset threshold, the region is marked as a candidate region. 3) Merge regions that satisfy any of the above rules into candidate variant regions.

4. The method for identifying genomic structural variations according to claim 1, characterized in that, The partial assembly in step S3 includes the following steps: 1) Extract Reads that meet the following conditions from the candidate mutation regions as seed Reads: Cover the candidate mutation region; It may contain soft-sheared fragments or split alignment information; 2) Based on the seed Reads, construct a local overlap map according to the overlap relationship between the Reads; 3) Perform path search and extension operations in the overlapping graph to generate at least one candidate contig sequence; 4) Perform consistency correction on the candidate contig sequences to obtain the target contig sequence; 5) Use the target contig sequence for subsequent breakpoint identification.

5. The method for identifying genomic structural variations according to claim 1, characterized in that, The multi-dimensional feature alignment model in step S5 includes the following steps: 1) Align the target contig sequence obtained in step 4) with the reference genome to obtain the initial alignment result; 2) Based on the alignment results, extract the following feature parameters: Sequence similarity features are used to characterize the degree of matching between the contig sequence and the reference sequence; Breakpoint location features are used to characterize the coordinates and distribution of discontinuous positions in the comparison. Coverage depth feature is used to characterize the number and distribution of Reads supporting the contig; Repeating structural features are used to characterize the length of repeating sequences and the characteristics of repeating units in the target region; 3) Normalize the above feature parameters to obtain standardized feature vectors; 4) Calculate the comprehensive alignment score based on the standardized feature vectors; 5) Based on the comprehensive alignment score and breakpoint location characteristics, determine the breakpoint location of the candidate structural variation region.

6. The method for identifying genomic structural variations according to claim 5, characterized in that, The sequence similarity features are calculated based on the ratio of matching bases or the alignment score; The breakpoint location features are determined based on the gap, soft shear location, or splitting alignment location in the comparison. The coverage depth feature is calculated based on the number of Reads supporting the region or the coverage uniformity. The repeating structural features are calculated based on the changes in the length and copy number of the repeating unit.

7. The method for identifying genomic structural variations according to claim 1, characterized in that, The calculation of the breakpoint support number in step S6 includes: 1) Based on the candidate breakpoint locations determined in step S5, extract all Reads and contig sequences that cross or are adjacent to the breakpoints; 2) Determine whether each Read or contig supports the breakpoint. The determination rules include: Reads are split and aligned at this position; Or Reads may contain soft shear structures that cross the break point; Or the contig sequence may have an alignment discontinuity at this position; 3) Count the number of Reads and contigs that satisfy the judgment rules, and use them as the support number of the breakpoint.

8. The method for identifying genomic structural variations according to claim 1, characterized in that, The rules for determining the type of structural variation include: When an extra sequence is inserted, it is considered an insertion mutation. When there are consecutive deletions in the reference sequence, it is determined to be a deletion variant; When the sequence direction is reversed, it is determined to be an inversion mutation; When the number of repeating units increases, it is determined to be a repeat amplification mutation.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the genome structural variation identification method as described in any one of claims 1-8.

Images