Training method and device of heterozygosity missing region detection model, equipment and medium
Patent Information
- Application Number
- CN202380016420.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-19
- Publication Date
- 2025-02-21
AI Technical Summary
In the prior art, the detection method for heterozygous missing areas is lengthy and consumes a lot of time and labor costs, resulting in inefficient detection and inability to popularize it.
By obtaining low-deep data in high-deep genome sequencing data, determining the information value and feature information of chromosomal regions, building an input feature matrix, and iteratively training the model using the self-attention allocation algorithm to obtain a model for detecting heterozygous missing region, and then Accurate detection of heterozygous missing areas under low-depth data.
The efficiency of heterozygous missing area detection is greatly improved, the dependence on high-deep sequencing data is reduced, the detection cost is reduced, and the rapid and accurate detection results are achieved.
Smart Images
Figure CN119497890A_ABST
Abstract
Description
Training method, device, equipment and medium for loss of heterozygosity region detection model Technical Field
[0001] The present disclosure generally relates to the field of computer technology, and more particularly to a method, apparatus, device, and medium for training a loss of heterozygosity region detection model. Background Art
[0002] Loss of heterozygosity (LOH) is a common form of genomic variation characterized by the loss or mutation of a single allele, resulting in a previously heterozygous locus becoming homozygous. This typically affects the entire gene and its adjacent chromosomal regions, and a large number of genetic diseases are caused by LOH regions within chromosomes.
[0003] Because the LOH region is essentially a region composed of multiple consecutive single nucleotide polymorphisms (SNPs), and the number of SNPs contained in different LOH regions varies, accurate detection of LOH regions requires testing high-depth genomic sequencing data. For example, the detection of LOH regions is completed through chromosome microarray (CMA) technology, which obtains a large amount of data information. Subsequently, different thresholds need to be set based on the characteristics of different LOH regions to screen out possible LOH regions. At the same time, it is necessary to consult a large amount of literature or databases based on the screened information to annotate the data and ultimately obtain a reasonable result report.
[0004] However, as a high-throughput and high-resolution screening technology, CMA obtains very large amounts of LOH information while ensuring data accuracy. As can be seen from the above process, it requires lengthy analysis steps, consumes a lot of time and manpower costs, and thus leads to low detection efficiency, making it impossible to popularize.
[0005] Summary of the Invention
[0006] In view of the above-mentioned defects or deficiencies in the prior art, it is desired to provide a training method, device, equipment and medium for a loss of heterozygosity region detection model, which can solve the problem that the current method for detecting LOH regions is lengthy and consumes a lot of time and manpower costs, and greatly improve detection efficiency.
[0007] In a first aspect, a method for training a loss of heterozygosity region detection model is provided, the method comprising:
[0008] Obtaining first data, where the first data is low-depth genome sequencing data obtained based on high-depth genome sequencing data, the first data including M types of low-depth genome sequencing data, the high-depth genome sequencing data including data of loss of heterozygosity regions, where M is a positive integer;
[0009] Determine an information value for each of N regions in the chromosome corresponding to the first data, wherein the information value includes mean value information and variance information for the regions, wherein the N regions include heterozygous regions and loss of heterozygosity regions, and N is a positive integer;
[0010] Determining feature information of the N regions, and determining an input feature matrix based on the feature information of the N regions and the information values of the N regions; the feature information includes an association between information of loss of heterozygosity regions corresponding to the N regions and the information values of each of the N regions;
[0011] The initial model is iteratively trained according to the above input feature matrix to obtain a model for detecting the loss of heterozygosity region. The above initial model is determined according to the self-attention allocation algorithm.
[0012] In the present disclosure, a group or several groups of different low-depth genome sequencing data (the high-depth genome sequencing data includes data of heterozygosity loss regions) obtained based on high-depth genome sequencing data are first obtained; then, the information values (average information and variance information) of the heterozygosity loss regions in all regions (i.e., N regions) in the chromosome corresponding to the above-mentioned first data are determined; thereafter, the correlation between the heterozygosity loss region information corresponding to the N regions and the information values of the N regions is determined, and the input feature matrix for training is determined based on the correlation between the above-mentioned heterozygosity loss region information and the information values of each of the above-mentioned regions; finally, the initial model is iteratively trained according to the input feature matrix to obtain a model for detecting heterozygosity loss regions. In this way, low-depth genome sequencing data is obtained from high-depth genome sequencing data of known heterozygosity loss regions and the information values and feature information of the above-mentioned low-depth genome sequencing data are determined, and then the training samples obtained, i.e., the input feature matrix, are used to train the model, and finally a heterozygosity loss region detection model is obtained. Therefore, in the subsequent use of this model for detecting loss of heterozygosity regions, low-depth genome sequencing data can be used to accurately measure the loss of heterozygosity regions in the chromosomes corresponding to the genome sequencing data. There is no need for high-depth genome sequencing data or other more complex high-throughput detection methods to accurately measure the loss of heterozygosity regions, thereby greatly improving the detection efficiency.
[0013] In a second aspect, a training device for a loss of heterozygosity region detection model is provided, the device comprising:
[0014] an acquisition module configured to acquire first data in a first format, wherein the first data is low-depth genome sequencing data acquired based on high-depth genome sequencing data, the first data comprising M types of low-depth genome sequencing data, the high-depth genome sequencing data including data of loss of heterozygosity regions, where M is a positive integer;
[0015] a determination module, configured to determine an information value of each of the N regions in the chromosome corresponding to the first data obtained by the acquisition module, wherein the information value includes mean value information and variance information of the regions, the N regions include heterozygous regions and loss of heterozygosity regions, and N is a positive integer;
[0016] The determination module is configured to determine characteristic information of the N regions and determine an input characteristic matrix based on the characteristic information of the N regions and the information values of the N regions; the characteristic information includes the correlation between the LOH region information corresponding to the N regions and the information value of each of the N regions;
[0017] The training module is used to iteratively train the initial model according to the input feature matrix determined by the determination module to obtain a heterozygous region detection model. The initial model is determined according to the self-attention allocation algorithm.
[0018] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method described in the first aspect is implemented.
[0019] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the program is executed by a processor, the method described in the first aspect is implemented.
[0020] In a fifth aspect, a computer program product is provided. The computer program product includes instructions, and when the instructions are executed by a processor, the method described in the first aspect is implemented.
[0021] Additional aspects and advantages of the present disclosure will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Other features, objects and advantages of the present disclosure will become more apparent from a reading of the detailed description of non-limiting embodiments made with reference to the following drawings:
[0023] FIG1 is a flow chart of a method for training a loss of heterozygosity region detection model according to an embodiment of the present disclosure;
[0024] FIG2 is a schematic diagram of the training process of the training method for the loss of heterozygosity region detection model provided by an embodiment of the present disclosure;
[0025] FIG3 is a schematic diagram of the structure of a training device for a loss of heterozygosity region detection model provided by an embodiment of the present disclosure;
[0026] FIG4 is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION
[0027] The present disclosure will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the invention are shown in the accompanying drawings.
[0028] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0029] The following are explanations of the professional terms that appear in this disclosure:
[0030] 1. LOH
[0031] LOH is a chromosomal event that can cause the loss of an entire gene and its surrounding chromosomal regions. If SNPs exist in the genomes of the father and mother, then the two alleles of the offspring (one from the father and one from the mother) exist in the region of these SNPs, and these regions are heterozygous (heterozygous). However, if the gene copy of one of the parents containing such a region is lost, there will be only one copy of this region, so this region has lost heterozygosity, that is, LOH. In short, if a gene copy from the father or mother is lost, the region with SNPs cannot show a heterozygous state.
[0032] LOH is loss of heterozygosity (LOH): it refers to the phenomenon that one of the two alleles in a specific region of chromosome is lost or mutated, resulting in only one allele being retained in the region.
[0033] 2. SNP
[0034] SNPs (Single Nucleotide Polymorphisms) refer to DNA sequence polymorphisms caused by variations in a single nucleotide at the genomic level, including single base transitions and transversions. SNP location classification can be categorized into three types: gene coding region, gene periphery, and intergenic SNPs. SNP location regions include UTRs, exons or introns of coding regions, and alternative splice sites.
[0035] The application scenarios of the present disclosure are introduced as follows:
[0036] Loss of heterozygosity (LOH) is a genomic variation characterized by the loss or mutation of one allele of a homologous chromosome pair, while the other allele remains intact. The genetic mechanism of LOH was first elucidated in 1929 through studies of X-ray-induced mutation sites in Drosophila melanogaster. LOH is prevalent in cancer and can lead to the inactivation of tumor suppressor genes, thereby promoting cancer development. Potential mechanisms of LOH include chromosome loss, partial chromosome loss, and gene conversion. In recent years, LOH has become a major research hotspot in genetic testing. It is now recognized that LOH is associated with many genetic diseases, including Mendelian recessive disorders. The presence of LOH on a chromosome can indicate the possibility of uniparental disomy (UPD), which occurs when both chromosomes in a homologous pair are inherited from the same parent. When UPD occurs on specific chromosomes, it can cause related diseases due to genetic imprinting. Numerous studies have shown that the risk of Mendelian recessive disorders is significantly increased in regions with LOH.
[0037] In recent years, with the continuous development of high-throughput sequencing technology, whole genome sequencing (WGS) has been widely used in disease prevention and treatment, such as genetic diseases, rare syndromes, and complex diseases. In clinical testing, due to the low cost and high accuracy of detecting copy number abnormalities, CNV-seq is widely used in low-depth WGS testing and plays a significant role in prenatal screening. However, while CNV-seq can detect copy number abnormalities and chromosome number abnormalities, it is less than ideal for detecting regions with loss of heterozygosity. This limitation is due to the generally low data depth of CNV-seq. When using this detection method, corresponding probes are set for high-frequency alleles in SNP regions within the genome. The method of detecting LOH based on pure heterozygosity of SNPs cannot accurately detect LOH regions, resulting in a large number of missed detections.
[0038] In addition, existing methods for detecting LOH regions also include: short tandem repeat (STR) detection method, methylation detection (MS-PCR / MS-MLPA) method, chromosome microarray (CMA) detection method, and bioinformatics software (PLINK) detection method.
[0039] The bioinformatics software is: PLINK uses an observational approach, employing a fixed-size sliding window to scan each chromosome for consecutive homozygous SNPs. PLINK first calculates the proportion of fully homozygous windows containing a particular SNP. If this proportion exceeds a pre-set threshold, the SNP is considered to be within a segment of LOH. A specific number of missing or heterozygous SNPs can be specified within each sliding window to account for genotyping errors, failures, or rare variants. Finally, if the number of consecutive homozygous SNPs within a segment exceeds a threshold (number of SNPs or chromosome distance), the segment is considered LOH.
[0040] Among the above methods, STR detection requires the selection of highly polymorphic STR markers based on the detection purpose and genomic location, which imposes certain limitations on the detection method and increases the detection cost. The MS-PCR / MS-MLPA detection method is time-consuming and therefore has a high time cost.
[0041] Currently, the most ideal technology for detecting LOH is the chromosome microarray (CMA) detection method. However, as a high-throughput and high-resolution screening technology, CMA obtains a very large amount of LOH information while ensuring data accuracy. Different thresholds need to be set for screening according to different purposes. At the same time, it is necessary to consult a large amount of literature or databases to annotate the data based on the screening information in order to finally obtain a reasonable result report.
[0042] The bioinformatics software PLINK lacks accuracy for clinical data, resulting in missed detections and false positives. This is because, when sequencing depth is too low, counting only SNP sites makes it impossible to determine whether the resulting SNPs are genuine, let alone LOH regions.
[0043] Therefore, a simple, fast and efficient method for detecting LOH areas is needed to solve the problems that current methods for detecting LOH areas are either lengthy, consuming a lot of time and manpower costs, or unable to accurately detect LOH areas.
[0044] Figure 1 is a flow chart of a method for training a model for detecting a region of heterozygosity loss according to an embodiment of the present disclosure. The method can be executed by a computer or other terminal. As shown in Figure 3, the method includes the following steps:
[0045] 301. Obtain first data.
[0046] In the embodiment of the present disclosure, the first data is low-depth genome sequencing data obtained based on high-depth genome sequencing data.
[0047] In the embodiment of the present disclosure, the first data includes M types of low-depth genome sequencing data, where M is a positive integer.
[0048] In the disclosed embodiment, the high-depth genome sequencing data includes data of heterozygosity loss regions.
[0049] It is understood that the depth mentioned above refers to sequencing depth. Generally, when measuring genomic sequencing data on chromosomes, the amount of genetic data on each chromosome is very large, so it is necessary to measure at a certain depth. Different measurement depths can be used for different measurement purposes or different measurement conditions.
[0050] Illustratively, the above-mentioned M types of low-depth genome sequencing data may be one type or multiple types.
[0051] In a possible embodiment, when there are multiple types of the above-mentioned M types of low-depth genome sequencing data, the above-mentioned M types of low-depth genome sequencing data may include any or all of 3X, 1X, 0.5X, and 0.1X genome sequencing data.
[0052] In a possible embodiment, when there are multiple types of the above-mentioned M types of low-depth genome sequencing data, the above-mentioned M types of low-depth genome sequencing data may include any one of 3X, 1X, 0.5X, and 0.1X genome sequencing data.
[0053] In the disclosed embodiment, the first data may be extracted from high-depth genome sequencing data. For example, the first data may include 3X, 1X, 0.5X, and 0.1X genome sequencing data, all of which are extracted from 30X high-depth genome sequencing data.
[0054] In the embodiment of the present disclosure, the first data is genome sequencing data in a first format.
[0055] In the embodiment of the present disclosure, the first format refers to the data format of the first data, for example, BAM format, fastq format, cram format, and sam format.
[0056] It is understandable that the above-mentioned high-depth genome sequencing data includes data on loss of heterozygosity regions. The reason is that it serves as the source data for subsequent training samples. Since the purpose of the present disclosure is to train a detection model that can accurately identify LOH regions, genome sequencing data with LOH regions must be used, and the information of LOH regions in the genome sequencing data (the location and size of the LOH regions, etc.) must be clear.
[0057] 302. Determine the information value of each of the N regions in the chromosome corresponding to the first data.
[0058] In the embodiment of the present disclosure, the above-mentioned information value includes the average value information and variance information of the above-mentioned regions, the above-mentioned N regions include heterozygous regions and heterozygous loss regions, and N is a positive integer.
[0059] In the disclosed embodiments, the N regions refer to observing the chromosome in a preset unit range, thereby dividing the chromosome corresponding to the first data into a plurality of regions, wherein the N regions include both heterozygous regions and loss of heterozygosity (LOH) regions. Furthermore, the loss of heterozygosity (LOH) regions are homozygous regions. It should be noted that the N regions, including heterozygous regions and LOH regions, are known before training begins.
[0060] For example, if the preset unit range is 10 KB, then 10 KB on the chromosome is used as an observation unit, and the loci on the chromosome corresponding to the first data within the 10 KB unit range are sequentially detected and observed, ultimately determining the loci of the SNP loci in N regions, including the N regions of the chromosome. The loci status includes, for example, whether the locus belongs to an LOH region or a non-LOH region.
[0061] In the embodiment of the present disclosure, after N regions are determined, the SNP sites in each of the N regions are calculated, and finally the mean value information and variance information of the SNP sites in each of the N regions are obtained.
[0062] In an example, the above-mentioned average value information may be an average value of the log-likelihood ratio, and the above-mentioned variance may be a variance of the log-likelihood ratio.
[0063] 303. Determine feature information of the N regions, and determine an input feature matrix according to the feature information of the N regions and information values of the N regions.
[0064] In the embodiment of the present disclosure, the feature information includes the LOH region information corresponding to the N regions and the association relationship between the information values of each of the N regions.
[0065] In the embodiment of the present disclosure, the LOH region information corresponding to the N regions refers to information about whether the center point of each of the N regions contains an identified LOH region.
[0066] In one example, the LOH region information corresponding to the N regions may be in a coded form, for example, 0 represents a region without LOH, and 1 represents a region with LOH.
[0067] In the embodiment of the present disclosure, the correlation relationship between the information values of each of the above-mentioned areas means that each of the N areas contains a large amount of average value information and variance information, and the relationship between the average value information and variance information in each area and between the areas needs to be observed and recorded, ultimately forming the correlation relationship between the information values of each of the N areas.
[0068] It is understandable that, as can be seen from the above content, the LOH region is actually composed of multiple homozygous SNP sites. On the contrary, the region where multiple homozygous SNP sites are located is not necessarily an LOH region, but may be a normal genetic phenomenon. The LOH region is specific. Therefore, in the previous technology of this disclosure, the accuracy of LOH region detection based on low-depth genome sequencing data is very low. Therefore, after determining the N regions where the SNP sites are located, the relationship between different SNP sites should be further analyzed, that is, it is necessary to analyze the correlation between the heterozygous region information corresponding to the above N regions and the information value of each region of the above N regions, so as to make the training sample information richer, so that when these samples are used to generate the final training samples for model training, the model can be fully trained, so that the model can be accurately detected when it is subsequently used to identify and detect LOH regions.
[0069] In the embodiment of the present disclosure, the above-mentioned method of determining the characteristic information may be a method of extracting the characteristic information using a sliding window, that is, the above-mentioned characteristic information may be extracted by sliding a sliding window on the chromosome corresponding to the first data.
[0070] Exemplarily, the sliding window may be preset or custom-set. For example, the sliding window may be 2.5 MB in size, and its sliding step may be consistent with the size of each of the N regions. For example, if the size of each region is 10 KB, the sliding step is 10 KB.
[0071] In the embodiment of the present disclosure, the above-mentioned input feature matrix is a training sample for training the LOH region detection model.
[0072] 304. Iteratively train the initial model based on the above input feature matrix to obtain a model for detecting loss of heterozygosity regions.
[0073] In the embodiment of the present disclosure, the above-mentioned initial model is determined based on the self-attention allocation algorithm.
[0074] In one embodiment, the initial model may be based on a bidirectional gated recurrent unit (BGRU) model.
[0075] It is understood that the self-attention allocation algorithm includes the self-attention mechanism and other algorithms based on self-attention extensions. Generally, the algorithm consists of an input layer, a self-attention layer, a normalization layer, a hidden layer, and an output layer of neurons. When the self-attention allocation algorithm processes the input feature matrix, the input feature matrix first passes through the input layer, then the self-attention layer, then the normalization layer, and then through the hidden layer with 64 neurons to the output layer of one neuron. After the output, one training is completed.
[0076] In the process of inputting the matrix, as shown in FIG2 , the input matrix is input row by row, and a single row input constitutes one training session, and the training is iterated back and forth until all training is completed.
[0077] Furthermore, in the above iterative training process, the activation function is the ReLU function, as shown in the following formula 1: f(x) = max(0,x) Formula 1
[0078] Exemplarily, the above-mentioned heterozygosity loss region model is a recurrent neural network (RNN) model, specifically, a bidirectional gated recurrent unit neural network model (BGRU) model.
[0079] Example 1: The following is a specific example to explain the above steps:
[0080] Assume that the first data includes 3X, 1X, 0.5X, and 0.1X genome sequencing data, and these genome sequencing data are in BAM format; the region size of each region in N regions is 10kb; the sliding window for extracting feature information is 2.5MB, and the sliding step size is 10kb:
[0081] First, the aforementioned 3X, 1X, 0.5X, and 0.1X genome sequencing data are obtained; then, the mean value information and variance information of the N regions containing SNP sites in the chromosome corresponding to the above four groups of genome sequencing data are determined; then, the correlation between the LOH region information corresponding to the N regions and the mean value information and variance information of the N regions is determined; then, the input feature matrix for training is determined using the correlation between the mean value information and variance information of the N regions, the LOH region information corresponding to the N regions, and the mean value information and variance information of the N regions; finally, the BGRU initial model is iteratively trained using the above input feature matrix to obtain the LOH region model detection model.
[0082] In the method provided by the embodiment of the present disclosure, a group or several groups of different low-depth genome sequencing data (the high-depth genome sequencing data includes data of heterozygosity loss regions) obtained based on high-depth genome sequencing data are first obtained; then, the information value (average information and variance information) of each region in the heterozygosity loss region of all regions (i.e., N regions) in the chromosome corresponding to the above-mentioned first data is determined; thereafter, the correlation between the heterozygosity loss region information corresponding to the N regions and the information value of the N regions is determined, and the input feature matrix for training is determined based on the correlation between the above-mentioned heterozygosity loss region information and the information value of each of the above-mentioned regions; finally, the initial model is iteratively trained according to the input feature matrix to obtain a model for detecting heterozygosity loss regions. In this way, low-depth genome sequencing data is obtained from high-depth genome sequencing data of known heterozygosity loss regions and the information value and feature information of the above-mentioned low-depth genome sequencing data are determined, and then the training sample obtained, i.e., the input feature matrix, is used to train the model, and finally a heterozygosity loss region detection model is obtained. In this way, in the subsequent use of this model for detecting loss of heterozygosity regions, low-depth genome sequencing data can be used to accurately measure the loss of heterozygosity regions in the chromosomes corresponding to the genome sequencing data. There is no need for high-depth genome sequencing data or other more complex high-throughput detection methods to accurately measure the loss of heterozygosity regions, thereby greatly improving the detection efficiency.
[0083] In another embodiment of the present disclosure, a specific implementation method for determining the information value of each of the N regions is also provided. Exemplarily, the specific implementation of "determining the information value of each of the N regions in the chromosome corresponding to the above-mentioned first data" mentioned above includes: calculating the likelihood value of each single nucleotide polymorphism site in the chromosome corresponding to the above-mentioned first data; calculating the log-likelihood information of each of the above-mentioned single nucleotide polymorphism sites based on the likelihood value of each of the above-mentioned single nucleotide polymorphism sites; dividing the above-mentioned chromosome corresponding to the above-mentioned first data into N regions according to a preset range, and determining the information value of each of the above-mentioned N regions based on the log-likelihood information of each of the above-mentioned single nucleotide polymorphism sites.
[0084] Illustratively, the aforementioned likelihood values include a first likelihood value of a haplotype and a second likelihood value of a diplotype.
[0085] Exemplarily, the above-mentioned calculation of the likelihood value of each single nucleotide polymorphism SNP site in the chromosome corresponding to the above-mentioned first data includes: calculating a first likelihood value of the haplotype of each SNP site in the chromosome corresponding to the above-mentioned first data and a second likelihood value of the diplotype of each SNP site in the chromosome corresponding to the above-mentioned first data.
[0086] Exemplarily, the above-mentioned SNP sites are known SNP sites. It should be noted that non-SNP sites do not participate in the above-mentioned calculation process.
[0087] In one example, the method for calculating the first likelihood value of the haplotype is as follows: Formula 2: Pmonosomy(A∧B)=f(AB) Formula 2
[0088] The method for calculating the second likelihood value of diploidy is as follows: Formula 3: Pdisomy(A∧B)=1 / 2f(AB)+1 / 2f(A)f(B) Formula 3
[0089] A and B above refer to alleles in the population, and f(A) and f(B) refer to the frequencies of alleles A and B in the population. A and B are SNP sites.
[0090] Exemplarily, the log-likelihood information may be a log-likelihood ratio.
[0091] Exemplarily, since the above likelihood values are divided into a first likelihood value of a haplotype and a second likelihood value of a diplotype, the log likelihood ratio of each SNP site is calculated based on the first likelihood value of the haplotype and the second likelihood value of the diplotype.
[0092] In one example, the above formula 4 for calculating the log-likelihood ratio of each SNP site is as follows:
[0093] Exemplarily, as mentioned above, the information value includes variance information and mean information.
[0094] For example, the preset range may be a preset range of data of a single calculated information value divided according to the amount that can be observed or calculated at a single time. For example, the preset range may be 10 kb.
[0095] It can be understood that after obtaining the aforementioned likelihood value and log-likelihood ratio, the chromosome is divided into N regions according to a preset range, and the average log-likelihood ratio and the variance of the log-likelihood ratio of all SNP sites in each of the N regions are calculated.
[0096] Furthermore, the above calculation method quantifies the probability that two sequencing reads originate from the same chromosome based on the known linkage disequilibrium (LD) patterns among the alleles observed in the reads. The allele frequencies and LD patterns of a large population (such as the 1000 Genomes sequencing data) reference panel are used to interpret potential relationships between sparse reads.
[0097] In another embodiment of the present disclosure, a specific implementation method for encoding the LOH region to facilitate the subsequent determination of the input feature matrix is also provided. For example, before the aforementioned "determining the feature information of the N regions, and determining the input feature matrix based on the feature information of the N regions and the information values of the N regions," the specific implementation also includes: dividing the N regions into X unit regions; detecting the center points of the X unit regions; and adding encoding information to each of the X unit regions based on the detection results.
[0098] Exemplarily, the X unit areas include at least one area among the N areas, and there is an overlapping area between the X unit areas.
[0099] Illustratively, the above-mentioned encoding information is used to indicate that each of the above-mentioned unit regions includes a heterozygous loss region.
[0100] For example, as can be seen from the above, the above method of determining feature information can be obtained through a sliding window. Therefore, the above detection of the center points of N regions can also be completed by a sliding window, for example, a sliding window with a size of 2.5MB and a sliding step size of 10kb.
[0101] Furthermore, the N regions are divided into X unit regions, and the size of each unit region can be consistent with the size of the sliding window. That is, the process of dividing the N regions can be completed by the sliding window.
[0102] Example 2: As shown in Figure 2, the sliding window size is 2.5 MB, and the sliding step size is 10 KB. Each sliding of the sliding window generates a unit area, and X sliding times generate X unit areas. The N unit areas overlap with each other and are evenly staggered by 10 KB.
[0103] For example, the above codes may be numbers or letters. In the case where the codes are numbers, 1 may be used to indicate that the unit area includes the LOH area, and 0 may be used to indicate that the unit area does not include the LOH area.
[0104] In this way, the LOH area in each unit area can be confirmed by dividing the unit area with overlapping parts and detecting the central location of the unit area, and coding information can be added to facilitate the subsequent generation of training samples, thereby obtaining a model that can confirm the LOH area based on low-depth data, thereby improving the efficiency of subsequent LOH detection.
[0105] In another embodiment of the present disclosure, it will be further explained how to generate the input feature matrix. For example, the above-mentioned "determining the feature information of the N regions, and determining the input feature matrix based on the feature information of the N regions and the information values of the N regions" includes: extracting the first feature information and the first information value corresponding to the N regions to construct a first input feature matrix, and extracting the second feature information and the second information value corresponding to the N regions to construct a second input feature matrix; merging the first input feature matrix and the second input feature matrix to generate the input feature matrix.
[0106] Exemplarily, the first characteristic information includes the correlation between the heterozygosity loss region information corresponding to the N regions and the average value of the log-likelihood ratio in the N regions, and the second characteristic information includes the correlation between the heterozygosity loss region information corresponding to the N regions and the variance of the log-likelihood ratio in the N regions.
[0107] Exemplarily, the first information value includes an average value of the log-likelihood ratios of the N regions, and the second information value includes a variance of the log-likelihood ratios of the N regions.
[0108] In an example, as described above, the above method of obtaining the first feature information of N regions and the second feature information of N regions can be obtained by sliding a sliding window on the chromosome.
[0109] Example 3: When each of the N regions is 10kb, the sliding window size is 2.5M, and the sliding step is 10KB, the sliding window is slid on the chromosome, as shown in Figure 2. Since the size of each window is 2.5MB, each time the sliding window slides, the average value of the log-likelihood ratios of the 250 regions, the correlation between the average values of the log-likelihood ratios of the 250 regions, and whether the center point of this sliding includes the coding information of the LOH region (i.e., the first information value and the first feature information) are obtained. At the same time, the variance of the log-likelihood ratios of the 250 regions, the correlation between the variances of the log-likelihood ratios of the 250 regions, and whether the center point of this sliding includes the coding information of the LOH region (i.e., the second information value and the second feature information) are also obtained.
[0110] Finally, after sliding X times, we get a first input feature matrix of X×1×250 (mean value matrix of log-likelihood ratio) and a second input feature matrix of X×1×250 (variance matrix of log-likelihood ratio).
[0111] Afterwards, the above two matrices are merged into a three-dimensional matrix, that is, the input feature matrix of X×2×250.
[0112] In this way, by integrating the eigenvalues, the relationships between them, and the LOH region information corresponding to N regions, we ultimately obtain a training sample: the input feature matrix. The model subsequently trained using this input feature matrix can accurately detect LOH regions using low-depth genomic sequencing data, improving LOH region detection efficiency.
[0113] In another embodiment of the present disclosure, a specific method for obtaining M groups of first data is also disclosed. For example, in the aforementioned "obtaining first data", the specific implementation includes: when obtaining second data, extracting M groups of first data in a first format from the second data.
[0114] Exemplarily, the second data is high-depth genome sequencing data, and the first data is low-depth genome sequencing data.
[0115] Exemplarily, the data volume of the first data is smaller than the data volume of the second data.
[0116] Illustratively, the second data may be second data in a second format.
[0117] Furthermore, the first format and the second format may be the same or different, and this embodiment of the present application does not limit this. For example, the second format may be fastq format, cram format, Bam format, or sam format.
[0118] In one example, the extraction of M groups of first data in the first format from the second data can be accomplished by using a Picard tool.
[0119] It should be noted that, generally, the Picard tool can be used to extract the first data from the second data. Further, when the first format and the second format are different, when the Picard tool is used to extract the first data from the second data, the second format of the data can be converted into the first format while the data is being extracted.
[0120] For example, the current second data is 30X genome sequencing data. At this time, the tool Picard is used to extract genome sequencing data of different depths from the 30X genome sequencing data in sequence, extracting four groups of data at 3X, 1X, 0.5X, and 0.1X respectively, and simultaneously converting the data from CRAM format to BAM format.
[0121] Another embodiment of the present disclosure also discloses a specific detection process after obtaining a model for detecting LOH regions. The method includes: receiving third data to be detected; detecting the third data using the LOH region detection model to determine LOH region detection information for the chromosomal locus corresponding to the third data; and confirming the LOH region result corresponding to the third data based on the detection information.
[0122] Exemplarily, the third data mentioned above is low-depth genome sequencing data.
[0123] Illustratively, the detection information is used to indicate the possibility of a heterozygosity loss region in the chromosome locus data corresponding to the third data.
[0124] It will be appreciated that, during the aforementioned training process, the initial model is trained using data information from LOH regions and data information from non-LOH regions to generate a loss of heterozygosity region detection model. Thus, when the loss of heterozygosity region detection model is tested on the third data, the model can detect LOH regions in the third data. Specifically, each locus corresponding to the third data can be tested, and then the probability value of each locus in the third data being located in the LOH region can be used to indicate whether the locus belongs to the LOH region.
[0125] Example 4: When the aforementioned LOH region information and non-LOH region information are represented by 0 or 1, after inputting the third data into the loss of heterozygosity region detection model, a decimal value in the range [0, 1] is output for each locus. The closer the value is to 1, the greater the probability that the point is located in an LOH region. When the output values for consecutive loci on the chromosome corresponding to the third data are all greater than 0.6, the region formed by these consecutive loci is an LOH region.
[0126] In another embodiment of the present disclosure, the specific process of confirming the result of the loss of heterozygosity region corresponding to the third data is also disclosed. In the aforementioned "confirming the result of the loss of heterozygosity region corresponding to the third data based on the above-mentioned detection information", the specific implementation includes: confirming the suspected loss of heterozygosity region in the chromosome corresponding to the third data based on the above-mentioned detection information; screening out the exposed Unmask region in the suspected loss of heterozygosity region to obtain the confirmed loss of heterozygosity region; merging the confirmed loss of heterozygosity regions to confirm the result of the loss of heterozygosity region corresponding to the third data.
[0127] It can be understood that in the above-mentioned Example 4, the region composed of continuous sites is actually a suspected LOH region. After obtaining the suspected LOH region, the positions of all LOH regions can be located. If the position distance between the suspected LOH regions on the same chromosome is less than 1MB, these regions are merged, and the unmasked region of the chromosome region is removed from the merged region.
[0128] It should be noted that if the unmasked area is located inside the LOH area, it is regarded as two LOH areas. Finally, the area larger than 3MB is determined to be the LOH area and the result is output.
[0129] The following embodiment of the present disclosure further describes the above solution using a practical example. The model generation process of the practical example may correspond to FIG2 :
[0130] As shown in Table 1, 10 high-depth genome sequencing data sets (i.e., the second data set) were used. The 30X depth data contained 10 LOH regions, totaling approximately 122.73 MB, and was formatted in CRAM format (i.e., the second format). The 30X genome CRAM data sets were grouped and extracted using the Picard tool into four data sets: 3X, 1X, 0.5X, and 0.1X, and then converted into BAM files.
[0131] Using the aforementioned formulas 2, 3, and 4, the mean and variance of the log-likelihood ratios of N regions within each 10KB range on each chromosome corresponding to the BAM file of each data set are obtained. Next, a sliding window with a size of 2.5MB and a step size of 10KB is used to slide on the chromosome to obtain the mean and variance of the log-likelihood ratios of 250 regions in each window to construct a 2×250 feature matrix, sliding the window Z times in total. At the same time, it is determined whether the center point of the sliding window is contained in the known LOH region, and the information of this heterozygous loss region is recorded together. Based on this information, all sites on the chromosome are divided into two categories: those that exist in the LOH region and those that do not exist in the LOH region. It is also necessary to record the correlation between different variances or different average values in each window.
[0132] Based on this information, a Z × 1 × 250 matrix is constructed using the correlation between the LOH region information corresponding to the N regions and the average log-likelihood ratios in the N regions, as well as the average log-likelihood ratios in the N regions. A Z × 1 × 250 matrix is also constructed using the correlation between the LOH region information corresponding to the N regions and the variances of the log-likelihood ratios in the N regions, as well as the variances of the log-likelihood ratios in the N regions. These two matrices are combined into a Z × 2 × 250 three-dimensional input feature matrix.
[0133] The above three-dimensional feature matrix is used as the model input matrix. A custom neural network is used to construct a loss of heterozygosity region detection model.
[0134] Table 1
[0135] After obtaining the above-mentioned model for detecting loss of heterozygosity regions, 306 samples (including 781 known LOH regions) were used to verify the accuracy of the model.
[0136] After the model is input to obtain suspected LOH regions, the positions of all LOH regions are located, and regions on the same chromosome that are less than 1MB apart are merged. The unmasked regions of the chromosome region are removed from the merged regions. If the unmasked region is located inside the LOH region, it is considered to be two LOH regions. The regions larger than 3MB are determined to be LOH regions and the results are output.
[0137] Finally, we define the region where the identified LOH region has 50% intersection with the known LOH region as the accurately identified region (refer to the following formula 4). At different depths of 3X, 1X, 0.5X, and 0.1X, 781, 781, 780, and 750 LOH regions were identified, respectively, and the recognition accuracy remained above 50%, as shown in Table 2.
[0138] Table 2
[0139] Among them, LOH_lengthCMA represents the length of the LOH region result obtained by the CMA experiment provided by the 1000 Genomes sequencing data, and LOH_lengthpredict represents the length of the LOH region result predicted by the model.
[0140] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the training rule determination method described in the present disclosure. For example, the steps of the method shown in FIG1 may be executed.
[0141] An embodiment of the present disclosure provides a computer program product. The computer program product includes instructions. When the instructions are executed by a processor, the steps of the method shown in FIG. 1 are implemented.
[0142] It should be noted that although the operations of the disclosed method are depicted in a particular order in the drawings, this does not require or imply that the operations must be performed in this particular order, or that all illustrated operations must be performed, to achieve desirable results.
[0143] FIG3 is a block diagram of a training device for a loss of heterozygosity region detection model according to an embodiment of the present disclosure. Referring to FIG3 , the device includes an acquisition module 601 , a determination module 602 , and a training module 603 .
[0144] An acquisition module 601 is configured to acquire first data, where the first data is low-depth genome sequencing data acquired based on high-depth genome sequencing data, the first data including M types of low-depth genome sequencing data, the high-depth genome sequencing data including data of loss of heterozygosity regions, where M is a positive integer;
[0145] a determination module 602 configured to determine an information value for each of the N regions in the chromosome corresponding to the first data obtained by the acquisition module 601, wherein the information value includes mean value information and variance information for the regions, wherein the N regions include heterozygous regions and loss of heterozygosity regions, and N is a positive integer;
[0146] The determination module 602 is configured to determine characteristic information of the N regions and determine an input characteristic matrix based on the characteristic information of the N regions and the information values of the N regions; the characteristic information includes the correlation between the LOH region information corresponding to the N regions and the information values of each of the N regions;
[0147] The training module 603 is used to iteratively train the initial model according to the input feature matrix determined by the determination module 602 to obtain a heterozygous region detection model. The initial model is determined according to the self-attention allocation algorithm.
[0148] In one embodiment, the apparatus includes a calculation module configured to: calculate a likelihood value of each single nucleotide polymorphism site in the chromosome corresponding to the first data, the likelihood value including a first likelihood value of a haplotype and a second likelihood value of a diplotype;
[0149] Calculating the log-likelihood information of each single nucleotide polymorphism site according to the likelihood value of each single nucleotide polymorphism site;
[0150] The determination module 602 is specifically configured to divide the chromosome corresponding to the first data into N regions according to a preset range, and determine the information value of each of the N regions based on the log-likelihood information of each single nucleotide polymorphism site.
[0151] In one embodiment, the calculation module is further used to calculate a first likelihood value of a haplotype of each single nucleotide polymorphism site in the chromosome corresponding to the first data and a second likelihood value of a diplotype of each single nucleotide polymorphism site in the chromosome corresponding to the first data.
[0152] In one embodiment, the determination module 602 is specifically configured to:
[0153] Divide the N areas into X unit areas, where each of the X unit areas includes at least one area in the N areas, and the X unit areas have overlapping areas;
[0154] Detect the center sites of the above X unit areas;
[0155] According to the detection result, coding information is added to each of the X unit regions, where the coding information is used to indicate that each of the unit regions includes a loss of heterozygosity region.
[0156] In one embodiment, the determination module 602 is specifically configured to:
[0157] Extracting first feature information and first information values corresponding to the N regions to construct a first input feature matrix, and extracting second feature information and second information values corresponding to the N regions to construct a second input feature matrix, wherein the first feature information includes a correlation between the loss of heterozygosity region information corresponding to the N regions and an average value of the log-likelihood ratios in the N regions, and the second feature information includes a correlation between the loss of heterozygosity region information corresponding to the N regions and a variance of the log-likelihood ratios in the N regions; the first information value includes the average value of the log-likelihood ratios of the N regions, and the second information value includes the variance of the log-likelihood ratios of the N regions;
[0158] The first input feature matrix and the second input feature matrix are combined to generate the input feature matrix.
[0159] In one embodiment, the acquisition module 601 is specifically used to: extract M groups of first data in the first format from the second data when acquiring the second data, wherein the second data is high-depth genome sequencing data and the first data is low-depth genome sequencing data.
[0160] In one embodiment, the above-mentioned device also includes: a receiving module for receiving third data to be detected, where the above-mentioned third data is low-depth genome sequencing data; a determination module 602 for detecting the above-mentioned third data using the above-mentioned heterozygosity loss region detection model, and determining the heterozygosity loss region detection information of the chromosome site corresponding to the above-mentioned third data; the above-mentioned detection information is used to indicate the possibility of the heterozygosity loss region of the chromosome site data corresponding to the above-mentioned third data; and based on the above-mentioned detection information, confirming the heterozygosity loss region result corresponding to the above-mentioned third data.
[0161] In one embodiment, the determining module 602 is further configured to:
[0162] Confirming the suspected loss of heterozygosity region in the chromosome corresponding to the third data based on the detection information;
[0163] Eliminate the Unmasked regions in the suspected LOH regions to obtain confirmed LOH regions;
[0164] According to the preset spacing range, the confirmed heterozygous loss regions are merged, and the heterozygous loss region results corresponding to the third data are confirmed.
[0165] In the embodiment of the present disclosure, the training device of the heterozygosity loss region detection model first obtains a group or several groups of different low-depth genome sequencing data obtained based on high-depth genome sequencing data (the high-depth genome sequencing data includes data on heterozygosity loss regions); then, the information value (average information and variance information) of the heterozygosity loss region in all regions (i.e., N regions) in the chromosome corresponding to the above-mentioned first data is determined; thereafter, the correlation between the heterozygosity loss region information corresponding to the N regions and the information value of the N regions is determined, and the input feature matrix for training is determined based on the correlation between the above-mentioned heterozygosity loss region information and the information value of each of the above-mentioned regions; finally, the initial model is iteratively trained according to the input feature matrix to obtain a model for detecting heterozygosity loss regions. In this way, low-depth genome sequencing data is obtained from the high-depth genome sequencing data of the known heterozygosity loss region and the information value and feature information of the low-depth genome sequencing data are determined, and then the training sample obtained, i.e., the input feature matrix, is used to train the model, and finally a heterozygosity loss region detection model is obtained. In this way, in the subsequent use of this model for detecting loss of heterozygosity regions, low-depth genome sequencing data can be used to accurately measure the loss of heterozygosity regions in the chromosomes corresponding to the genome sequencing data. There is no need for high-depth genome sequencing data or other more complex high-throughput detection methods to accurately measure the loss of heterozygosity regions, thereby greatly improving the detection efficiency.
[0166] It should be understood that the units recorded in the training device of the heterozygosity loss region detection model correspond to the various steps in the method described in the accompanying drawings. Therefore, the operations and features described above for the method are also applicable to the training device of the heterozygosity loss region detection model, the resource access device and the units contained therein, and will not be repeated here. The training device and resource access device of the heterozygosity loss region detection model can be pre-implemented in the browser or other security applications of the computer device, or can be loaded into the browser or its security application of the computer device by downloading or other means. The corresponding units in the training device and resource access device of the heterozygosity loss region detection model can cooperate with the units in the computer device to implement the solution of the embodiment of the present disclosure.
[0167] The several modules or units mentioned in the detailed description above are not necessarily divided into one module or unit. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.
[0168] It should be noted that for details not disclosed in the training device and resource access device of the heterozygosity loss region detection model in the embodiments of the present disclosure, please refer to the details disclosed in the above embodiments of the present disclosure, and no further details will be given here.
[0169] 4, FIG4 shows a block diagram of a computer device suitable for implementing an embodiment of the present disclosure. As shown in FIG4, a computer system 1700 includes a central processing unit (CPU) 1701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1702 or a program loaded from a storage portion 1708 into a random access memory (RAM) 1703. In RAM 1703, various programs and data required for the operating instructions of the system are also stored. CPU 1701, ROM 1702, and RAM 1703 are connected to each other via a bus 1704. An input / output (I / O) interface 1705 is also connected to bus 1704.
[0170] The following components are connected to the I / O interface 1705: an input section 1706 including a keyboard, a mouse, and the like; an output section 1707 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 1708 including devices such as a hard disk; and a communication section 1709 including a network interface card such as a LAN card or a modem. The communication section 1709 performs communication processing via a network such as the Internet. A drive 1710 is also connected to the I / O interface 1705 as needed. Removable media 1711, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 1710 as needed, so that computer programs read therefrom can be installed into the storage section 1708 as needed.
[0171] In particular, according to an embodiment of the present disclosure, the process described above with reference to flowchart Figure 1 can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1709, and / or installed from a removable medium 1711. When the computer program is executed by the central processing unit (CPU) 1701, the above-mentioned functions defined in the system of the present disclosure are executed.
[0172] It should be noted that the computer-readable medium described in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical cable, RF, or any suitable combination thereof.
[0173] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operating instructions of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the aforementioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than the order marked in the accompanying drawings. For example, the boxes represented by two connections can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operating instruction, or can be implemented using a combination of dedicated hardware and computer instructions.
[0174] The units or modules described in the embodiments of this disclosure may be implemented in software or hardware. The units or modules described may also be provided in a processor. For example, a processor may be described as including a first receiving module, a second receiving module, and a sending module. The names of these units or modules do not, in some cases, limit the units or modules themselves.
[0175] As another aspect, the present disclosure further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or may exist independently without being incorporated into the electronic device. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the training rule determination method and the black market business identification method described in the present disclosure.
[0176] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the aforementioned disclosed concepts. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.
Claims
1. A method for training a heterozygous region detection model, comprising: Acquire first data, where the first data is low-depth genome sequencing data acquired based on high-depth genome sequencing data, the first data includes M types of low-depth genome sequencing data, the high-depth genome sequencing data includes data of heterozygosity loss regions, and M is a positive integer; Determine the information value of each region in N regions in the chromosome corresponding to the first data, wherein the information value includes mean value information and variance information of the region, the N regions include heterozygous regions and heterozygous loss regions, and N is a positive integer; Determine feature information of the N regions, and determine an input feature matrix according to the feature information of the N regions and information values of the N regions; The characteristic information includes the association relationship between the heterozygous loss region information corresponding to the N regions and the information value of each region of the N regions; The initial model is iteratively trained according to the input feature matrix to obtain a heterozygous loss region detection model, wherein the initial model is determined according to a self-attention allocation algorithm.
2. The method according to claim 1, wherein: The determining the information value of each of the N regions in the chromosome corresponding to the first data includes: Calculating the likelihood value of each single nucleotide polymorphism site in the chromosome corresponding to the first data, wherein the likelihood value includes a first likelihood value of a haplotype and a second likelihood value of a diplotype; Calculate the log-likelihood information of each single nucleotide polymorphism site according to the likelihood value of each single nucleotide polymorphism site; The chromosome corresponding to the first data is divided into N regions according to a preset range, and the information value of each region in the N regions is determined according to the log-likelihood information of each single nucleotide polymorphism site.
3. The method according to claim 2, wherein: The calculating the likelihood value of each single nucleotide polymorphism site in the chromosome corresponding to the first data comprises: A first likelihood value of a haplotype of each single nucleotide polymorphism site in the chromosome corresponding to the first data and a second likelihood value of a diplotype of each single nucleotide polymorphism site in the chromosome corresponding to the first data are calculated.
4. The method according to claim 1, wherein: Before determining the feature information of the N regions and determining the input feature matrix according to the feature information of the N regions and the information values of the N regions, the method further includes: The N regions are divided into X unit regions, each of which includes at least one of the N regions. There are overlapping areas between X unit areas; Detecting the center sites of the X unit regions; According to the detection result, coding information is added to each of the X unit regions, where the coding information is used to indicate that each unit region includes a loss of heterozygosity region.
5. The method according to claim 1, wherein: The determining the feature information of the N regions, and determining the input feature matrix according to the feature information of the N regions and the information values of the N regions, comprises: Extracting first feature information and first information values corresponding to the N regions to construct a first input feature matrix, and extracting second feature information and second information values corresponding to the N regions to construct a second input feature matrix, wherein the first feature information includes the correlation between heterozygosity loss region information corresponding to the N regions and the average value of the log-likelihood ratio in the N regions, and the second feature information includes the correlation between the heterozygosity loss region information corresponding to the N regions and the variance of the log-likelihood ratio in the N regions; the first information value includes the average value of the log-likelihood ratio of the N regions, and the second information value includes the variance of the log-likelihood ratio of the N regions; The first input feature matrix and the second input feature matrix are combined to generate the input feature matrix.
6. The method according to claim 1, wherein: The obtaining of the first data includes: When obtaining second data, M groups of first data are extracted from the second data, where the second data is high-depth genome sequencing data and the first data is low-depth genome sequencing data.
7. The method according to claim 1, wherein: The method further comprises: Receiving third data to be detected, where the third data is low-depth genome sequencing data; Detecting the third data using the heterozygosity loss region detection model to determine heterozygosity loss region detection information of the chromosome locus corresponding to the third data; the detection information is used to indicate the possibility of the heterozygosity loss region of the chromosome locus data corresponding to the third data; According to the detection information, the heterozygous loss region result corresponding to the third data is confirmed.
8. The method according to claim 7, wherein: The step of confirming the heterozygous region result corresponding to the third data according to the detection information includes: According to the detection information, confirming the suspected heterozygous loss region in the chromosome corresponding to the third data; Screening out exposed regions in the suspected heterozygous loss region to obtain a confirmed heterozygous loss region; According to a preset spacing range, the confirmed heterozygous loss regions are merged, and the heterozygous loss region results corresponding to the third data are confirmed.
9. A training device for a heterozygous region detection model, comprising: An acquisition module is used to acquire first data, wherein the first data is low-depth genome sequencing data acquired based on high-depth genome sequencing data, and the first data includes M types of low-depth genome sequencing data. The high-depth genome sequencing data includes data of heterozygosity loss regions, and M is a positive integer; A determination module, used to determine the information value of each region in the N regions of the chromosome corresponding to the first data acquired by the acquisition module, wherein the information value includes the mean value information and the variance information of the region, the N regions include heterozygous regions and heterozygous loss regions, and N is a positive integer; The determination module is used to determine the feature information of the N regions, and determine the input feature matrix according to the feature information of the N regions and the information values of the N regions; The characteristic information includes the association relationship between the heterozygous loss region information corresponding to the N regions and the information value of each region of the N regions; The training module is used to iteratively train the initial model according to the input feature matrix determined by the determination module to obtain a heterozygous loss region detection model, and the initial model is determined according to a self-attention allocation algorithm.
10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 8 is implemented.
11. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.
12. A computer program product, comprising instructions, wherein the instructions are executed by a processor to implement the method according to any one of claims 1 to 8.