A method, system, medium and electronic device for detecting a homozygous region of a gene

Abstract

The application discloses a kind of detection methods, systems, media and electronic equipment of homozygous region of gene, comprising: obtaining SNP site and corresponding allele frequency, based on sliding window strategy, feature extraction is carried out to the SNP site, obtain several characteristic data, the characteristic data is divided into heterozygous site density feature, allele frequency offset feature, sequencing depth consistency feature and region continuity feature according to feature type;The genome is divided into several candidate homozygous regions according to the preset region segmentation strategy;The characteristic data is sent to machine learning model, and homozygous region probability score is obtained;The homozygous region probability score is verified, and a detection report is generated;Realize full-scale homozygous region detection from 300kb to whole chromosome range, significantly better than BAF threshold method can only detect greater than 5Mb event, realize accurate identification.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics detection technology, and in particular to a method, system, medium, and electronic device for detecting homozygous gene regions. Background Technology

[0002] Spontaneous miscarriage is one of the most common complications of pregnancy. Approximately 50%–60% of early miscarriages are caused by chromosomal abnormalities (aneuploidy) in the embryo, while a significant proportion of the remaining cases involve chromosomal structural abnormalities or genomic copy number variations. In recent years, with the development of high-throughput sequencing technologies, chromosomal microarray analysis (CMA) and low-depth whole-genome sequencing (CNV-seq) have become routine methods for the genetic testing of miscarriage products, effectively identifying chromosomal aneuploidy and large-segment copy number variations. However, these technologies have limited ability to detect homozygous regions of the genome, especially loss of heterozygosity (LOH) and uniparental disomy (UPD).

[0003] Existing methods identify homozygous regions by detecting the degree to which the B-allele frequency (BAF) deviates from 0.5. However, the resolution of this method is usually limited to the megabase (Mb) level, making it difficult to detect fine events at the kilobase (kb) level, which may involve key imprinted genes or recessive pathogenic genes. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method, system, medium, and electronic device for detecting homozygous gene regions.

[0005] To achieve the above objectives, the present invention employs the following technical solution: a method for detecting homozygous gene regions, comprising: SNP sites and their corresponding allele frequencies are obtained, wherein the SNP sites are obtained by preprocessing sequencing data obtained from genome sequencing; Based on the sliding window strategy, features are extracted from the SNP sites to obtain several feature data. The feature data are classified into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to feature type. The genome is divided into several candidate homozygous regions according to a preset region segmentation strategy; The feature data is sent to a machine learning model to obtain a homozygous region probability score. The probability score of the homozygous region is verified, and a detection report is generated.

[0006] As a further description of the above technical solution: Preprocessing of sequencing data includes: The sequencing data underwent adapter removal and low-quality base trimming to obtain an optimized dataset. The optimized dataset is compared with a preset human reference genome to identify the SNP sites and determine the allele frequency of each SNP site.

[0007] As a further description of the above technical solution: the formula for calculating the heterozygous site density characteristic is as follows: ; in, Display window Density of heterozygous sites within; Display window The number of internal heterozygous sites; Display window The actual length; The formula for calculating the allele frequency shift feature is as follows: ; in, and These represent the sequencing depths of the reference allele and the alternative allele, respectively. Allele frequency; Within the sliding window, calculate the median allele frequency for all heterozygous loci. ; Determine allele frequency shift offset The larger the value, the more the region deviates from the normal heterozygous state, and the more likely it is to be a homozygous region; The formula for calculating the sequencing depth consistency feature is as follows: ; in, This represents the coefficient of variation per kilobase sequencing depth within the window. The mean coefficient of variation for the entire genome; The regional continuity feature To calculate the heterozygous site density characteristics of the current window and several adjacent windows. The mean of the Spearman correlation coefficients with allele frequency shift feature B was obtained.

[0008] As a further description of the above technical solution: the step of dividing the genome into several candidate homozygous regions according to a preset region segmentation strategy includes: The chromosomes corresponding to the genome are initialized as candidate blocks; The heterozygous site density difference between adjacent sub-regions is calculated by sliding within the candidate block, wherein the sub-regions are divided based on the candidate block length using a region segmentation strategy. The region segmentation strategy is as follows: for a sub-region whose length is greater than a first preset threshold, if the difference in heterozygous site density is greater than a second preset threshold, the sub-region is segmented at the peak of the difference. For sub-regions whose length is less than the first preset threshold, the similarity with adjacent sub-regions is calculated. If the similarity is greater than the third preset threshold, the sub-regions are merged until all the sub-regions meet the second preset threshold or the third preset threshold. The sub-regions are then output as the candidate homozygous regions.

[0009] As a further description of the above technical solution: it also includes training the machine learning model, including: Construct a training dataset and a validation set, and then standardize the dataset. The parameters of the machine learning model are updated using a cross-validation strategy, and the dataset is input into the machine learning model to obtain the homozygous region probability score. The optimal classification threshold is determined based on the validation set as the fourth preset threshold. When the probability score of the homozygous region is greater than the fourth preset threshold, it is determined to be a homozygous region.

[0010] As a further description of the above technical solution: the machine learning model is a gradient boosting decision tree classification model, and the parameters include decision tree depth, learning rate and subsampling ratio. The decision tree depth ranges from 3 to 7, the learning rate ranges from 0.01 to 0.3, and the subsampling ratio ranges from 0.6 to 0.9.

[0011] As a further description of the above technical solution: the verification of the homozygous region probability score includes: If the homozygous region on a single chromosome exceeds 90% and the total length of the chromosome is greater than 40 megabases, it is determined to be a whole chromosome uniparental diploid. Determination of large fragment homozygosity: If the length of the homozygous region is more than 10 to 40 megabases and does not cross the centromere, it is determined to be a large fragment loss of heterozygosity. Determination of intermediate segment homozygosity: If the length of the homozygous region is greater than 2 but less than 10 megabases, it is determined to be an intermediate segment heterozygous deletion. Fine fragment homozygosity determination: If the length of the homozygous region is greater than 0.3 but less than 2 megabases, it is determined to be a fine fragment heterozygosity deletion; In addition, when making the determination, false positive detection is performed on homozygous regions. The false positive detection technology includes excluding regions located in highly repetitive sequence regions, regions with sequencing depth less than 30% of the median depth of the whole genome, or regions with a frequency greater than 5% in the control sample library.

[0012] It also includes a system for detecting homozygous gene regions, comprising: The acquisition module obtains SNP sites and their corresponding allele frequencies, wherein the SNP sites are obtained by preprocessing the sequencing data obtained from genome sequencing. The feature extraction module extracts features from the SNP sites based on a sliding window strategy, resulting in several feature data. These feature data are categorized into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to their feature types. The region segmentation module is used to divide the genome into several candidate homozygous regions according to a preset region segmentation strategy; The processing module sends the feature data to the machine learning model to obtain a homozygous region probability score; The detection module verifies the probability score of the homozygous region and generates a detection report.

[0013] It also includes a computer-readable storage medium storing a computer program for running the detection method, wherein the computer program causes a computer to perform the detection method as described in any of the above technical solutions.

[0014] It also includes an electronic device, comprising: One or more processors; memory; and One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs being used to perform the detection method as described in any of the above technical solutions.

[0015] The above technical solution has the following advantages or beneficial effects: By processing the heterozygous site density of chromosomes, and dividing the region based on the heterozygous site density, several candidate homozygous regions are obtained. The probability score of homozygous regions is obtained through feature data and machine learning model, and the candidate homozygous regions are judged. This achieves full-scale homozygous region detection from 300kb to the entire chromosome range, which is significantly better than the BAF threshold method, which can only detect events larger than 5Mb (megabases), and achieves accurate identification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the detection method proposed in this invention; Figure 2 This is a flowchart of the sequencing data preprocessing process in this invention; Figure 3 This is a flowchart illustrating the process of dividing the genome into several candidate homozygous regions in this invention; Figure 4 This is a flowchart of the process of training a machine learning model in this invention; Figure 5 This is a schematic diagram of the detection system proposed in this invention. Figure 6 This is a schematic diagram illustrating the principle of dividing the genome into several candidate homozygous regions in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0019] In obstetric clinical practice, approximately 40%–50% of early miscarriages show "chromosomal normality" according to traditional karyotype analysis or chromosome microarray analysis (CMA), with the genetic cause remaining unclear for a long time. With further research, homozygous regions of the genome, particularly UPD and LOH involving imprinted genes, have been confirmed as important pathogenic factors. Homozygous events in imprinted gene regions (such as 11p15.5, 15q11-q13, etc.) can lead to serious imprinting disorders such as Beckwith-Wiedemann syndrome and Angelman syndrome, which are closely related to abnormal embryonic development and miscarriage. However, current detection technologies, such as BAF, are limited to the megabase (Mb) level and cannot detect fine events at the kilobase (kb) level. Furthermore, BAF does not integrate multi-dimensional evidence such as heterozygous site density and sequencing depth consistency, resulting in insufficient specificity. Threshold settings are often empirically fixed values, failing to consider chromosomal region characteristics (such as naturally low heterozygosity near telomeres / centromeres) and sample specificity (such as sequencing depth fluctuations), making it difficult to control false positive rates while maintaining high sensitivity.

[0020] Reference Figure 1 One embodiment of the present invention provides a method for detecting homozygous gene regions, comprising: S1. Obtain SNP sites and their corresponding allele frequencies. The SNP sites are obtained by preprocessing the sequencing data obtained from genome sequencing. S2. Based on the sliding window strategy, feature extraction is performed on SNP sites to obtain several feature data. The feature data are divided into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to feature type. S3. Divide the genome into several candidate homozygous regions according to a preset region segmentation strategy; S4. Send the feature data to the machine learning model to obtain the homozygous region probability score; S5. Verify the probability score of homozygous regions and generate a test report.

[0021] In this embodiment, the genome sequencing data is preprocessed to obtain an optimized dataset. This optimized dataset is then compared with the human reference genome to identify SNP (single nucleotide polymorphism) sites and their corresponding allele frequencies. A sliding window strategy is used to extract features from the genome to obtain heterozygous site density characteristics. Four types of data: A) allele frequency offset, B) sequencing depth consistency, C) regional continuity, and R) The genome is divided into multiple continuous segments with uniform internal features and clear boundaries (B, C, R). Regions with significantly reduced heterozygosity and obvious allele shifts are selected as candidate homozygous regions for high-resolution localization. The four feature data are sent to a machine learning model to score each candidate homozygous region, outputting a homozygous region probability score between 0 and 1. A higher homozygous region probability score indicates a greater likelihood that the region is truly homozygous. Based on the homozygous region probability score and region length, a classification is performed to distinguish between UPD (uniparental diploid), large-fragment LOH, medium-fragment LOH, and fine-fragment LOH. False positive detection is also performed. Finally, a standardized test report is generated, including region location, length, type, clinical significance, and reliability. This method achieves full-scale homozygous region detection from 300kb (kilobases) to the entire chromosome, significantly outperforming the BAF threshold method, which can only detect events larger than 5Mb (megabases), thus achieving accurate identification.

[0022] Reference Figure 2 Preprocessing of sequencing data includes: S11. Adapter removal and low-quality base trimming are performed on the sequencing data to obtain an optimized dataset. S12. The optimized dataset is compared with the preset human reference genome to identify SNP sites and determine the allele frequency of each SNP site.

[0023] In this embodiment, whole-genome sequencing (≥30×) was performed on genomic DNA extracted from aborted fetal tissue samples to obtain the original FASTQ file, which is used to store biological sequences and their sequencing quality assessment data. Optimized datasets were obtained by adapter removal and low-quality base trimming using the Trimmomatic tool. The optimized datasets were then aligned to the human reference genome (GRCh38 / hg38) using the BWA-MEM algorithm. Post-alignment processing was performed using SAMtools and GATK tools, including sorting, deduplication, and local re-alignment. Finally, single nucleotide polymorphism (SNP) sites were identified using the bcftools mpileup-call workflow to obtain the allele frequency (AF) for each site. To improve the reliability of aborted fetal sample detection, strict quality control standards were set: sequencing data Q30 ≥ 85%, alignment rate ≥ 95%, effective sequencing depth ≥ 0.08× (CNV-seq) or ≥ 25× (WGS), and maternal cell contamination rate < 15% (assessed through sex chromosome analysis or SNP heterozygosity).

[0024] The formula for calculating the heterozygous site density characteristics is as follows: ; in, Display window Density of heterozygous sites within; Display window The number of internal heterozygous sites; Display window The actual length; The formula for calculating allele frequency shift characteristics is: ; in, and These represent the sequencing depths of the reference allele and the alternative allele, respectively. Allele frequency; Within the sliding window, calculate the median allele frequency for all heterozygous loci. ; Determine allele frequency shift offset The larger the value, the more the region deviates from the normal heterozygous state, and the more likely it is to be a homozygous region; The formula for calculating sequencing depth consistency characteristics is: ; in, This represents the coefficient of variation per kilobase sequencing depth within the window. The mean coefficient of variation for the entire genome; Regional continuity characteristics To calculate the heterozygous site density characteristics of the current window and several adjacent windows. The mean of the Spearman correlation coefficients with allele frequency shift feature B was obtained.

[0025] In this embodiment, the heterozygous site density characteristic reflects the enrichment degree of heterozygous SNP sites within the window; wherein For window Internal satisfaction The number of heterozygous sites, The actual window length (base pairs) is dynamically adjusted based on the characteristics of the chromosome region. The default window size for autosomal regions is 50 kilobases, while the window size near telomeres and centromeres is reduced to 20 kilobases. Allele frequency shift characteristics quantify the degree to which the overall allele frequency within the window deviates from 0.5. Sequencing depth consistency characteristics assess the stability of sequencing depth within the window and exclude data fluctuation interference. The coefficient of variation is the genome-wide average; a value closer to 1 indicates greater depth stability. Regional continuity is a measure of the similarity between a window and its neighboring windows, calculated by assessing the heterozygous site density between the current window and the three windows upstream and downstream. The Spearman mean correlation coefficient between the feature B and the allele frequency shift was obtained, reflecting the clarity of the homozygous region boundary. All feature data were Z-score normalized before being input into the subsequent machine learning model to eliminate dimensional differences between different features.

[0026] Reference Figure 3 and Figure 6 The genome is divided into several candidate homozygous regions according to a pre-defined region segmentation strategy, including: S31. Initialize the chromosomes corresponding to the genome as candidate blocks; S32. Calculate the heterozygous site density difference between adjacent sub-regions by sliding within the candidate block, where the sub-regions are divided based on the candidate block length using a region segmentation strategy.

[0027] The region segmentation strategy is as follows: for sub-regions with a length greater than a first preset threshold, if the difference in heterozygous site density is greater than a second preset threshold, the sub-regions are segmented at the peak of the difference. For sub-regions whose length is less than the first preset threshold, calculate the similarity with adjacent sub-regions. If the similarity is greater than the third preset threshold, merge the sub-regions until all sub-regions meet the second or third preset threshold, and output the sub-regions as candidate homozygous regions.

[0028] In this embodiment, the entire chromosome corresponding to the genome is initialized as a single candidate block and segmented to form several sub-regions. The length of each sub-region is then determined. Sub-regions larger than a first preset threshold, where the first preset threshold is the minimum region length. The default value is 100 kilobases, and the difference in heterozygous site density between adjacent subregions is calculated. The second preset threshold The segmentation threshold is... Greater than the second preset threshold The default value is 0.8. Segmentation is performed at the peak of the difference. For sub-regions whose length is less than the first preset threshold, the similarity with neighboring sub-regions is calculated. The algorithm formula is: ; Third preset threshold The merging threshold is set if the similarity exceeds the third preset threshold. If the default value is 0.85, then sub-region merging is performed. The above steps are repeated until all sub-regions formed by the final segmentation meet the second or third preset threshold. The segmented sub-regions are then output as candidate homozygous regions. A smaller window (20kb) is used in the telomere and centromere regions to improve the local resolution to the 300 kilobase level.

[0029] Furthermore, filtering can be performed based on feature data and output sub-regions, such as retaining those that satisfy the heterozygous site density characteristics of the sub-region. Heterozygous site density characteristics smaller than the whole genome And the allele frequency shift in the subregion The sub-regions are selected as candidate homozygous regions.

[0030] Reference Figure 4 It also includes training machine learning models, including: A1. Construct the training dataset and validation set, and standardize the dataset. A2. By using a cross-validation strategy, update the parameters of the machine learning model and input the dataset into the machine learning model to obtain the homozygous region probability score; A3. Determine the optimal classification threshold based on the validation set as the fourth preset threshold. When the probability score of a homozygous region is greater than the fourth preset threshold, it is determined to be a homozygous region.

[0031] In this embodiment, 150 abortion product samples verified by cytogenetics were collected. Among them, 78 positive samples containing known homozygous regions were selected as the dataset, including 22 samples containing the entire chromosome (UPD), 30 samples containing large LOH fragments, 18 samples containing medium LOH fragments, and 8 samples containing fine LOH fragments. 72 samples were normal diploids used as the negative control validation set. The dataset was standardized: Z-score standardization was performed on the four feature data of each sample to eliminate dimensional differences. The machine learning model was a gradient boosting decision tree (GBDT) classification model built with XGBoost, with parameters including decision tree depth, learning rate, and subsampling ratio. The feature data ( (B, C, R) and sub-region length A 5-fold cross-validation strategy is employed, optimizing the decision tree depth, learning rate, and subsampling ratio. The decision tree depth ranges from 3 to 7, the learning rate from 0.01 to 0.3, and the subsampling ratio from 0.6 to 0.9. A suitable subsampling ratio is a tree depth of 5, a learning rate of 0.2, a subsampling ratio of 0.7, and a feature subsampling ratio of 0.8. The algorithm iterates for 200 rounds on an independent validation set, based on the Youden index (J = Sensitivity + Specificity). 1. Select the optimal classification threshold As the fourth preset threshold, ranging from 0.65 to 0.75, when the homozygous region probability score... It was identified as a homozygous region, achieving a sensitivity of 96.3% and a specificity of 94.7%, with a detection rate of 89.2% for fine fragment events of 0.5–2 Mb.

[0032] Validation of the homozygous region probability score includes: If the homozygous region on a single chromosome exceeds 90% and the total length of the chromosome is greater than 40 megabases, it is determined to be a whole chromosome uniparental diploid. Determination of large fragment homozygosity: If the length of the homozygous region is more than 10 to 40 megabases and does not cross the centromere, it is determined to be a large fragment loss of heterozygosity. Determination of intermediate segment homozygosity: If the length of the homozygous region is greater than 2 but less than 10 megabases, it is determined to be an intermediate segment heterozygous deletion. Fine fragment homozygosity determination: If the length of the homozygous region is greater than 0.3 but less than 2 megabases, it is determined to be a fine fragment heterozygosity deletion; In addition, when making the determination, false positive detection is performed on homozygous regions. Technical false positive detection includes excluding regions located in highly repetitive sequence regions, regions with sequencing depth less than 30% of the median depth of the whole genome, or regions with a frequency greater than 5% in the control sample library.

[0033] In this embodiment, the probability score of homozygous regions is validated. The identified homozygous regions are validated through hierarchical verification. First, homozygous regions on the entire chromosome are validated, and then they are identified based on different length segments, which are respectively verified as large segment loss of heterozygosity, medium segment loss of heterozygosity, and fine segment loss of heterozygosity. Among them, fine segment loss of heterozygosity can effectively identify small homozygous events involving key imprinted gene clusters, such as 11p15.5, 15q11-q13, or recessive pathogenic genes. By querying imprinted gene databases, such as GeneImprint and OMIM, homozygous events involving imprinted regions or recessive pathogenic genes are marked, and a structured report with genetic counseling value is directly output. Pathogenic and benign homozygous events are distinguished to achieve accurate identification. The entire process analysis takes less than 2 hours (30×WGS data), supports batch processing, and is suitable for routine clinical laboratory testing.

[0034] Simultaneously, false positive detection is required to exclude regions located in highly repetitive regions, such as centromeres and telomeres, regions with sequencing depth less than 30% of the median depth of the whole genome, and regions that appear more than 5% of the frequency in the control sample library.

[0035] In one specific embodiment I. Clinical Sample Information Collection This embodiment included tissue samples of miscarriage products from 36 women of childbearing age with clear records of pregnancy outcomes, divided into two groups: Case group (n=18): a history of ≥2 early spontaneous abortions confirmed by ultrasound (gestation ≤10 weeks), and the most recent abortion product was confirmed by chromosome microarray analysis (CMA) to have a normal karyotype (46,XX or 46,XY, without pathogenic copy number variations). Control group (n=18): Healthy women who successfully conceived to full term (≥37 weeks) and had no history of miscarriage, whose miscarriage was due to planned termination of pregnancy (medical indication: fetal structural abnormalities confirmed by prenatal diagnosis, but normal chromosomal karyotype).

[0036] All samples were pathologically confirmed as chorionic villus tissue, and the maternal cell contamination rate was assessed by sex chromosome SNP heterozygosity, all of which were <15%. All participants signed informed consent forms, and the research protocol was approved by the local ethics committee and complies with the ethical guidelines of the Declaration of Helsinki.

[0037] Table 1 presents the baseline characteristics of the sample cohort, showcasing the clinical demographic characteristics of the 36 samples. There were no significant differences between the two groups in baseline characteristics such as age, preconception BMI, and gestational age (P>0.05), ensuring the comparability of subsequent genetic analyses.

[0038] Clinical features Case group (n=18) Control group (n=18) p-value Age (years) 0.62 Median (range) 32.5（28–39） 31.0（27–38） Preconception BMI (kg / m²) 0.41 Median (range) 22.8（19.5–27.3） 22.1（18.9–26.7） Gravidity <0.001 Median (range) 3.0（2–5） 2.0（1–3） Parity <0.001 Median (range) 0（0–1） 1（1–2） Number of miscarriages — — Median (range) 2.5（2–4） — Most recent pregnancy outcome — Early spontaneous abortion (≤10 weeks) 18（100%） — Full-term live birth (≥37 weeks) — 18（100%） CMA results of the abortion products — Normal karyotype (euploid) 18（100%） 18（100%） Sample type — villous tissue 18（100%） 18（100%） Mother source pollution rate (%) 0.38 Median (range) 8.7（5.2–14.3） 7.9（4.8–13.6） Table 1 Note: Parity refers to the total number of pregnancies, and lot number refers to the number of live births; maternal contamination rate is assessed using the frequency of heterozygous X chromosome loci. II. Sequencing Data Preprocessing Sequencing protocol: DNA source: In the case group, DNA was obtained from chorionic villus tissue after miscarriage (STR testing ruled out maternal contamination), while in the control group, it was obtained from peripheral blood. All samples underwent deep whole-genome sequencing (30×) on the Illumina NovaSeq 6000 platform. DNA extraction was performed using the QIAamp DNA Micro Kit (Qiagen), and library construction was performed using the KAPA HyperPrep Kit. Quality control standards: sequencing data Q30 ≥ 85%, alignment rate ≥ 95%, effective sequencing depth ≥ 25×, and maternal cell contamination rate < 15%.

[0039] Bioinformatics workflow: 1. Quality Control: FastQC assesses the quality of the raw data, Trimmomatic (v0.39) removes linkers and low-quality bases to obtain an optimized data set (parameters: LEADING:20 TRAILING:20 SLIDINGWINDOW:4:25 MINLEN:35). 2. Alignment: BWA-MEM (v0.7.17) aligned the optimized dataset to the GRCh38 reference genome with the parameter "-MR'@RG\tID:sample\tSM:sample\tPL:ILLUMINA'" to perform genome sequence alignment; 3. Post-processing: SAMtools (v1.10) for sorting and indexing, Picard (v2.23.8) for marking repeating sequences, and GATK (v4.2.0.0) for BaseRecalibrator and ApplyBQSR; 4. Mutation identification: GATK HaplotypeCaller identifies SNP sites in GVCF mode and performs joint genotyping by GenotypeGVCFs. Hard filtering retains high-quality SNP sites with QUAL>30, QD>2.0, FS<60.0, and MQ>40.0. 5. Allele frequency calculation: For each SNP locus, Depth is obtained through SAMtools depth.

[0040] III. Region Segmentation Strategy The goal of the region segmentation strategy is to divide the genome into contiguous segments with internal homogeneity, providing structured input for subsequent machine learning classification. By segmenting and merging sub-regions, the algorithm dynamically identifies mutation points with heterozygous site density, effectively distinguishing between truly homozygous regions and technical noise. Specifically targeting the fine structural characteristics of imprinted gene regions, the algorithm uses a smaller window (20kb) in the telomere / centromere region, improving local resolution to the 300kb level.

[0041] Minimum area length The optimal value was determined to be 100 kilobases through grid search optimization; blocks with values ​​lower than this are more susceptible to random fluctuations. Second preset threshold : Default 0.8 (after standardization), ensuring that segmentation only occurs at significant feature mutation points; Third preset threshold : Default 0.85, balances block fragmentation and excessive merging; Sliding window strategy: 50kb window for autosomal regions (step size 10kb), 20kb window for regions within 5Mb of telomeres / centromeres (step size 5kb) to improve imprinting region resolution.

[0042] IV. Machine Learning Models Input features include: heterozygous site density features Allele frequency shift feature B, sequencing depth consistency feature C Regional continuity characteristic R, sub-region length .

[0043] All feature data were Z-score standardized before being input into the machine learning model. Parameter settings: tree depth 5, learning rate 0.2, subsampling ratio 0.7, feature subsampling ratio 0.8, 200 iterations. Five-fold cross-validation was used, with the validation set as the optimization target. On an independent dataset (45 examples), the model achieved: AUC=0.973 (95%CI:0.948–0.991) Sensitivity = 96.3% (95% CI: 91.2–98.7%) Specificity = 94.7% (95% CI: 89.5–97.6%) Precision = 93.8% (95% CI: 88.4–97.0%) Stratified evaluation by fragment length showed that the detection rate of fine fragment events in the range of 0.5–2 Mb reached 89.2%, which was significantly better than the traditional BAF threshold method (52.4%).

[0044] V. Verification of the probability score for homozygous regions To verify the detection capability of this invention for key clinically significant imprinted regions, 12 clinically significant imprinted gene clusters (covering regions such as 11p15.5, 15q11-q13, and 14q32) were selected and systematically evaluated in a validation cohort of 36 cases. The comparison method was PennCNV (v1.0.5), a commonly used SNP chip-based LOH detection software in clinical practice, but its resolution is typically limited to events above 5 Mb.

[0045] Comparison of Imprint Area Detection Performance: Table 2 shows a comparison of the detection performance of the method in this embodiment and the conventional method (PennCNV) in 12 key imprinted regions. The results show that this embodiment maintains high sensitivity (85.7%–100%) in all imprinted regions, especially for fine fragment events <2Mb, such as the H19 / IGF2 imprinted control region at 11p15.5, where the detection rate reaches 92.3%, while PennCNV cannot identify such events at all (sensitivity 0%). In terms of specificity, this embodiment effectively filters false positives through multi-scale biological validation rules, achieving a specificity of 94.1%–100%, significantly better than PennCNV (76.5%–94.1%).

[0046]

[0047] Table 2 Wherein, sensitivity = true positive / (true positive + false negative); specificity = true negative / (true negative + false positive); precision = true positive / (true positive + false positive); "—" indicates that precision cannot be calculated due to no positive predictions. Case Study: 11p15.5 Fine Fragment UPD Detection The patient was a 32-year-old female, gravida 3, par 0, with a history of two early spontaneous abortions (7 weeks and 9 weeks of gestation). The aborted fetus was 46,XX as determined by CMA, with a normal karyotype. High-resolution LOH / UPD analysis was performed using the method of this invention: Key findings: A 0.52 Mb homozygous fragment (chr11: 2,016,406-2,536,406) was detected in the p15.5 region of the short arm of chromosome 11, covering the H19 / IGF2 imprinted control region (ICR1) and the upstream regulatory region of the CDKN1C gene; Machine learning score: 0.89 (threshold τ=0.68); Fragment characteristics: heterozygous site density. (Genome-wide mean 185), allele frequency shift Regional continuity ; Clinical interpretation: This region is a key imprinted region of Beckwith-Wiedemann syndrome. Maternal UPD11p15.5 can lead to IGF2 overexpression and CDKN1C expression loss, which is associated with abnormal embryonic development and miscarriage. Verification: Targeted sequencing confirmed that it is maternally derived UPD11p15.5. In contrast, PennCNV did not report any LOH events in this sample because its algorithm requires a minimum detection fragment length of 5Mb and cannot identify fine fragment events of 0.52Mb.

[0048] False positive detection technology: In the 36-case validation cohort, the method in this embodiment generated 3 false positive regions (all located in highly repetitive sequence regions, such as near the centromere). These were successfully filtered out using multi-scale biological validation rules (excluding highly repetitive regions, low-depth regions, and high-frequency polymorphic regions in the population), resulting in a final false positive rate of 0%. In contrast, PennCNV generated 7 false positives, 3 of which failed to be excluded by its built-in filtering rules (false positive rate of 8.3%). The main reason for this was its insufficient robustness to sequencing depth fluctuations and repetitive sequence regions.

[0049] VI. Comparative Analysis with Traditional Methods (PennCNV) Resolution comparison: This embodiment achieves a detection resolution of 300kb through a sliding window strategy and a region segmentation strategy; while PennCNV, based on a fixed window (typically ≥5Mb) and the BAF threshold method, has a resolution limited to above 5Mb. In a cohort of 36 cases, this embodiment detected 21 homozygous regions <5Mb (14 of which were <2Mb), while PennCNV only detected 5 events >5Mb, resulting in a false negative rate of 76.2%.

[0050] Robustness to parental contamination: Abortion product samples often contain 5%–15% maternal cell contamination, affecting the accuracy of LOH detection. In simulated contamination experiments (0%, 5%, 10%, and 15% contamination rates), the method of this invention maintained 89.2% sensitivity and 91.5% specificity at a 15% contamination rate; while PennCNV's sensitivity dropped to 68.3% at a 10% contamination rate and further to 52.1% at 15%, indicating that this invention is more adaptable to the characteristics of abortion product samples.

[0051] Clinical applicability comparison: Analysis time: The entire analysis process in this embodiment took 1 hour and 47 minutes (30×WGS data), while PennCNV requires 2 hours and 15 minutes (including data format conversion). Report content: This embodiment directly outputs a structured clinical report, annotating imprinted genes, pathogenicity rating, and genetic counseling recommendations; PennCNV only outputs a BED format region list, which requires manual interpretation. Applicable scenarios: This embodiment is specifically optimized for aborted tissue and supports low-purity samples; PennCNV is mainly designed for blood / saliva samples and has poor adaptability to tissue degradation samples.

[0052] As demonstrated by the above embodiments, this embodiment constructs an end-to-end, high-resolution, high-precision, and robust system for detecting homozygous genomic regions. While maintaining high sensitivity, it significantly reduces the false positive rate, making it particularly suitable for detecting LOH / UPD in aborted fetal tissue, ranging from hundreds of kilobases to entire chromosomes. Compared to traditional tools such as PennCNV, this embodiment exhibits significant advantages at clinically critical sites such as imprinted gene regions, providing a powerful tool for the genetic etiological diagnosis of unexplained abortion.

[0053] Reference Figure 5 It also includes a system for detecting homozygous gene regions, comprising: Acquisition module 1 acquires SNP sites and their corresponding allele frequencies. The SNP sites are obtained by preprocessing the sequencing data obtained from genome sequencing. Feature extraction module 2 extracts features from SNP sites based on a sliding window strategy, resulting in several feature data. The feature data are categorized into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to feature type. The region segmentation module 3 is used to divide the genome into several candidate homozygous regions according to a preset region segmentation strategy; Processing module 4 sends the feature data to the machine learning model to obtain the homozygous region probability score; Detection module 5 verifies the probability score of homozygous regions and generates a detection report.

[0054] In this embodiment, the acquisition module 1 acquires SNP sites and their corresponding allele frequencies and sends them to the feature extraction module 2. Through feature extraction, four types of feature data are obtained: heterozygous site density features, allele frequency offset features, sequencing depth consistency features, and regional continuity features. The region segmentation module 3 segments the genome corresponding to the SNP sites acquired by the acquisition module 1 to form several candidate homozygous regions. The processing module 4 acquires the candidate homozygous regions obtained by the region segmentation module 3 and the feature data obtained by the feature extraction module 2. Through machine learning model processing, a homozygous region probability score is obtained for each candidate homozygous region. Regions determined to be homozygous are sent to the detection module 5 for verification, and a corresponding detection report is output.

[0055] It also includes a computer-readable storage medium storing a computer program for running the detection method, wherein the computer program causes the computer to perform the following steps: S1. Obtain SNP sites and their corresponding allele frequencies. The SNP sites are obtained by preprocessing the sequencing data obtained from genome sequencing. S2. Based on the sliding window strategy, feature extraction is performed on SNP sites to obtain several feature data. The feature data are divided into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to feature type. S3. Divide the genome into several candidate homozygous regions according to a preset region segmentation strategy; S4. Send the feature data to the machine learning model to obtain the homozygous region probability score; S5. Verify the probability score of homozygous regions and generate a test report.

[0056] The computer-readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of a computer program from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the computer-readable storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the ASIC can reside in a user equipment. Of course, the processor and the computer-readable storage medium can also exist as discrete components in a communication device.

[0057] Specifically, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0058] It also includes an electronic device, comprising: One or more processors; memory; and One or more programs, wherein the programs are stored in memory and configured to be executed by one or more processors, the programs being used to perform the following steps: S1. Obtain SNP sites and their corresponding allele frequencies. The SNP sites are obtained by preprocessing the sequencing data obtained from genome sequencing. S2. Based on the sliding window strategy, feature extraction is performed on SNP sites to obtain several feature data. The feature data are divided into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to feature type. S3. Divide the genome into several candidate homozygous regions according to a preset region segmentation strategy; S4. Send the feature data to the machine learning model to obtain the homozygous region probability score; S5. Verify the probability score of homozygous regions and generate a test report.

[0059] Memory is used to store computer programs. This memory may include high-speed random access memory (RAM) and may also include non-volatile memory (Non-volatile memory). Volatile Memory (NVM), such as at least one disk storage device, can also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0060] A processor is used to execute computer programs stored in memory. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0061] Alternatively, the memory can be either standalone or integrated with the processor.

[0062] When memory is a device independent of the processor, electronic devices may also include a bus. This bus is used to connect the memory and the processor. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc.

[0063] It should be noted that, through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the prior art, can be embodied in the form of software products. These computer software products can be stored in computer-readable storage media, such as ROM / RAM, magnetic disks, optical disks, etc., and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or certain portions of the embodiments. In this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0064] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting homozygous gene regions, characterized in that, include: SNP sites and their corresponding allele frequencies are obtained, wherein the SNP sites are obtained by preprocessing sequencing data obtained from genome sequencing; Based on the sliding window strategy, features are extracted from the SNP sites to obtain several feature data. The feature data are classified into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to feature type. The genome is divided into several candidate homozygous regions according to a preset region segmentation strategy; The feature data is sent to a machine learning model to obtain a homozygous region probability score. The probability score of the homozygous region is verified, and a detection report is generated.

2. The detection method according to claim 1, characterized in that: Preprocessing of sequencing data includes: The sequencing data underwent adapter removal and low-quality base trimming to obtain an optimized dataset. The optimized dataset is compared with a preset human reference genome to identify the SNP sites and determine the allele frequency of each SNP site.

3. The detection method according to claim 1, characterized in that: The formula for calculating the heterozygous site density characteristics is as follows: ; in, Display window Density of heterozygous sites within; Display window The number of internal heterozygous sites; Display window The actual length; The formula for calculating the allele frequency shift feature is as follows: ; in, and These represent the sequencing depths of the reference allele and the alternative allele, respectively. Allele frequency; Within the sliding window, calculate the median allele frequency for all heterozygous loci. ; Determine allele frequency shift offset The larger the value, the more the region deviates from the normal heterozygous state, and the more likely it is to be a homozygous region; The formula for calculating the sequencing depth consistency feature is as follows: ; in, This represents the coefficient of variation per kilobase sequencing depth within the window. The mean coefficient of variation for the entire genome; The regional continuity feature To calculate the heterozygous site density characteristics of the current window and several adjacent windows. The mean of the Spearman correlation coefficients with allele frequency shift feature B was obtained.

4. The detection method according to claim 1, characterized in that: The step of dividing the genome into several candidate homozygous regions according to a preset region segmentation strategy includes: The chromosomes corresponding to the genome are initialized as candidate blocks; The heterozygous site density difference between adjacent sub-regions is calculated by sliding within the candidate block, wherein the sub-regions are divided based on the candidate block length using a region segmentation strategy. The region segmentation strategy is as follows: for a sub-region whose length is greater than a first preset threshold, if the difference in heterozygous site density is greater than a second preset threshold, the sub-region is segmented at the peak of the difference. For sub-regions whose length is less than the first preset threshold, the similarity with adjacent sub-regions is calculated. If the similarity is greater than the third preset threshold, the sub-regions are merged until all the sub-regions meet the second preset threshold or the third preset threshold. The sub-regions are then output as the candidate homozygous regions.

5. The detection method according to claim 1, characterized in that: It also includes training the machine learning model, including: Construct a training dataset and a validation set, and then standardize the dataset. The parameters of the machine learning model are updated using a cross-validation strategy, and the dataset is input into the machine learning model to obtain the homozygous region probability score. The optimal classification threshold is determined based on the validation set as the fourth preset threshold. When the probability score of the homozygous region is greater than the fourth preset threshold, it is determined to be a homozygous region.

6. The detection method according to claim 1, characterized in that: The machine learning model is a gradient boosting decision tree classification model, with parameters including decision tree depth, learning rate, and subsampling ratio. The decision tree depth ranges from 3 to 7, the learning rate ranges from 0.01 to 0.3, and the subsampling ratio ranges from 0.6 to 0.

9.

7. The detection method according to claim 1, characterized in that: The verification of the probability score of the homozygous region includes: If the homozygous region on a single chromosome exceeds 90% and the total length of the chromosome is greater than 40 megabases, it is determined to be a whole chromosome uniparental diploid. Determination of large fragment homozygosity: If the length of the homozygous region is more than 10 to 40 megabases and does not cross the centromere, it is determined to be a large fragment loss of heterozygosity. Determination of intermediate segment homozygosity: If the length of the homozygous region is greater than 2 but less than 10 megabases, it is determined to be an intermediate segment heterozygous deletion. Fine fragment homozygosity determination: If the length of the homozygous region is greater than 0.3 but less than 2 megabases, it is determined to be a fine fragment heterozygosity deletion; During the determination process, a false positive detection technique is also applied to homozygous regions. The false positive detection technology includes excluding regions located in highly repetitive sequence regions, regions with sequencing depth less than 30% of the median depth of the whole genome, or regions with a frequency greater than 5% in the control sample library.

8. A system for detecting homozygous gene regions, characterized in that, include: The acquisition module obtains SNP sites and their corresponding allele frequencies, wherein the SNP sites are obtained by preprocessing the sequencing data obtained from genome sequencing. The feature extraction module extracts features from the SNP sites based on a sliding window strategy, resulting in several feature data. These feature data are categorized into heterozygous site density features, allele frequency shift features, sequencing depth consistency features, and regional continuity features according to their feature types. The region segmentation module is used to divide the genome into several candidate homozygous regions according to a preset region segmentation strategy; The processing module sends the feature data to the machine learning model to obtain a homozygous region probability score; The detection module verifies the probability score of the homozygous region and generates a detection report.

9. A computer-readable storage medium, characterized in that, It stores a computer program for running a detection method, wherein the computer program causes a computer to perform the detection method as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: One or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs being used to perform the detection method as described in any one of claims 1-7.

Images