Gene data processing method, system, electronic device and storage medium

Abstract

The embodiments of the present application provide a method, system, electronic device and storage medium for processing genetic data, which belongs to the field of genetic analysis technology. The method obtains the genotype data of the individual to be tested, and determines multiple associated genes of the individual to be tested based on the genotype data and the single nucleotide polymorphism data associated with the target risk task; queries the associated gene features corresponding to each associated gene in the pre-constructed gene feature library, and the gene feature library includes embedded representations of different genes constructed based on the embedding model; the embedding model is obtained by self-supervised learning of multiple gene expression profiles of the same species as the individual to be tested; integrates the features of multiple associated gene features to obtain individual features; calls the prediction model to process the individual features to obtain the risk score corresponding to the individual to be tested and the target risk task. This method can improve the accuracy of genetic data processing.

Description

Technical Field

[0001] The present disclosure relates to the field of gene analysis technology, and in particular to a gene data processing method, system, electronic device, and storage medium. Background Art

[0002] The Polygenic Risk Score (PRS) model is a tool for assessing an individual's risk of developing a certain disease or trait. It integrates genetic variation information from multiple gene loci to evaluate an individual's genetic susceptibility. The model is currently widely used in scenarios such as disease risk prediction, clinical decision support, and drug development.

[0003] The PRS model is based on the results of genome-wide association studies (GWAS) to identify numerous single nucleotide polymorphisms (SNPs) in individuals that are associated with specific diseases or traits, and then integrates the effect values ​​of these related SNPs to calculate a risk score.

[0004] However, in related technologies, the method of using the PRS model for risk scoring only relies on the cumulative effect of a large number of SNPs for risk scoring. The cumulative effect of SNPs is difficult to intuitively link with biological pathways or functions, which limits the in-depth understanding of the risk mechanism and leads to low accuracy in predicting risk scores. Summary of the Invention

[0005] The main purpose of the embodiments of the present application is to propose a method, system, electronic device and storage medium for processing genetic data, aiming to improve the accuracy of risk score prediction for target risk tasks.

[0006] To achieve the above objectives, a first aspect of an embodiment of the present application provides a method for processing genetic data, the method comprising:

[0007] Acquiring genotype data of an individual to be tested, and determining a plurality of associated genes of the individual to be tested based on the genotype data and single nucleotide polymorphism data associated with the target risk task;

[0008] querying a pre-constructed gene signature library for each associated gene, wherein the gene signature library includes embedded representations of different genes based on an embedding model; the embedding model is obtained by self-supervised learning of multiple gene expression profiles of the same species as the individual to be tested;

[0009] Performing feature integration on the multiple associated gene features to obtain individual features;

[0010] The prediction model is called to process the individual characteristics to obtain a risk score corresponding to the individual to be tested and the target risk task, wherein the prediction model is trained based on the sample characteristics and corresponding phenotypic information of multiple sample individuals.

[0011] To achieve the above objectives, a second aspect of the embodiments of the present application provides a genetic data processing system, the system comprising:

[0012] a data acquisition unit, configured to acquire genotype data of the individual to be tested, and determine a plurality of associated genes of the individual to be tested based on the genotype data and single nucleotide polymorphism data associated with the target risk task;

[0013] a query unit, configured to query a pre-constructed feature library for associated gene features corresponding to each associated gene, wherein the gene feature library includes embedded representations of different genes constructed based on an embedding model; the embedding model is obtained by self-supervised learning of multiple gene expression profiles of the same species as the individual to be tested;

[0014] an integration unit, configured to integrate the plurality of associated gene features to obtain individual features;

[0015] The prediction unit is used to call the prediction model to process the individual characteristics to obtain the risk score corresponding to the individual to be tested and the target risk task, and the prediction model is trained based on the sample characteristics and corresponding phenotypic information of multiple sample individuals.

[0016] In some embodiments, the data acquisition unit includes:

[0017] an acquisition subunit, configured to acquire first single nucleotide polymorphism data associated with a target risk task;

[0018] a screening subunit, configured to screen the first single nucleotide polymorphism data to obtain a plurality of second single nucleotide polymorphism data based on a statistical test result of an allele frequency difference between a risk individual and a reference individual of the target risk task;

[0019] The determination subunit is used to determine a plurality of third single nucleotide polymorphism data based on the genotype data, and to determine a plurality of associated genes of the individual to be tested according to the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data.

[0020] Optionally, in some embodiments, the determining subunit includes:

[0021] An acquisition module, configured to acquire genomic positions and functional association information corresponding to the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data;

[0022] A mapping module is used to map the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data to corresponding genes according to the genome positions and the functional association information to obtain a plurality of associated genes.

[0023] Optionally, in some embodiments, the present application further provides a gene signature library construction device, comprising:

[0024] an acquiring unit, configured to acquire a plurality of first training samples, wherein the first training samples include gene expression profiles of human samples;

[0025] A training unit, configured to perform self-supervised learning on the embedding model using the first training sample;

[0026] A construction unit is used to construct a gene feature library based on the embedding representations of different genes extracted by the trained embedding model when the embedding model training converges.

[0027] Optionally, in some embodiments, the training unit includes:

[0028] a normalization subunit, configured to perform normalization processing on the gene expression profile of each of the first training samples, and determine a target training sample based on the normalization processing result;

[0029] The training subunit is used to obtain the embedding model by performing self-supervised learning using a random mask based on the target training sample.

[0030] Optionally, in some embodiments, the integration unit includes:

[0031] A second acquisition subunit is used to obtain weight coefficients corresponding to the multiple associated gene features;

[0032] The calculation subunit is used to perform weighted calculation on the multiple associated gene features based on the weight coefficient to obtain individual features.

[0033] Optionally, in some embodiments, the integration unit includes:

[0034] A convolution subunit, configured to perform convolution processing on the feature sequence composed of the plurality of associated gene features to obtain sequence features;

[0035] The integration subunit is used to integrate the multiple related gene features based on the sequence features to obtain individual features.

[0036] Optionally, in some embodiments, the present application further provides a model training device, comprising:

[0037] A third acquisition subunit is configured to acquire a plurality of second training samples, wherein the second training samples include sample genotype data of a plurality of sample individuals and corresponding phenotypic information labels;

[0038] a search subunit, configured to determine the sample-associated gene of each sample individual based on the sample genotype data, and search the gene signature corresponding to each sample-associated gene in the gene signature library to obtain a plurality of sample-associated gene signatures;

[0039] A prediction subunit, configured to integrate the plurality of sample-associated gene features to obtain sample individual features, and input the sample individual features into a prediction model to be trained to perform disease risk prediction and obtain a predicted risk score;

[0040] An updating subunit is configured to calculate a loss value according to the predicted risk score and the corresponding phenotypic information label, and to update the parameters of the prediction model based on the loss value.

[0041] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the genetic data processing method described in the first aspect when executing the computer program.

[0042] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the genetic data processing method described in the first aspect.

[0043] To achieve the above-mentioned objectives, the fifth aspect of an embodiment of the present application proposes a computer program product, which includes a computer program, and the computer program is read and executed by a processor of a computer device, so that the computer device executes the genetic data processing method described in the first aspect.

[0044] The method for processing genetic data proposed in the embodiment of the present application obtains the genotype data of the individual to be tested, and determines multiple associated genes of the individual to be tested based on the genotype data and single nucleotide polymorphism data associated with the target risk task; queries the associated gene features corresponding to each associated gene in a pre-constructed gene feature library, and the gene feature library includes embedded representations of different genes constructed based on an embedding model; the embedding model is obtained by self-supervised learning of multiple gene expression profiles of the same species as the individual to be tested; feature integration is performed on multiple associated gene features to obtain individual features; and a prediction model is called to process the individual features to obtain a risk score corresponding to the individual to be tested and the target risk task, and the prediction model is trained based on the sample features and corresponding phenotypic information of multiple sample individuals.

[0045] It can be seen from this that the method for processing genetic data provided in the embodiment of the present application, through a pre-constructed gene embedding feature library, enables the generation of individual embedded features based on the individual's genotype information and the gene embedding feature library when predicting the risk score of the target risk task for the individual. Further, the individual embedded features are predicted based on the prediction model obtained by supervised training to obtain a risk score. Since the method based on gene embedding integrated features captures the complex relationship between genes, it can better reflect the real biological effects than the simple SNP accumulation, and therefore can improve the accuracy of risk score prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide a further understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0047] Figure 1 A schematic diagram of the process of processing genetic data provided in this application;

[0048] Figure 2 A schematic diagram of the process for constructing a gene signature library provided in this application;

[0049] Figure 3 A flowchart of the model training method provided in this application;

[0050] Figure 4 A schematic diagram of the area under the ROC curve of the risk score prediction model provided in this application in an independent test set of simulated data;

[0051] Figure 5 Schematic diagram of the F1 score of the risk score prediction model provided in this application in an independent test set of simulated data;

[0052] Figure 6 Schematic diagram of the confusion matrix of the risk score prediction model provided in this application in an independent test set of simulated data;

[0053] Figure 7 Schematic diagram for evaluating the importance of features in each dimension of gene vector;

[0054] Figure 8a Schematic diagram of ROC comparison of the SNP scoring model in the solution provided in this application and related technologies in an independent test set;

[0055] Figure 8b Schematic diagram of the comparison of F1 scores of SNP-based scoring models in independent test sets in the solutions and related technologies provided in this application;

[0056] Figure 9is a schematic diagram of the structure of the genetic data processing system provided in an embodiment of the present application;

[0057] Figure 10 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0059] Before further explaining the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations:

[0060] Transformer model: A deep learning model based on the attention mechanism, originally used for machine translation tasks, and now widely used in natural language processing, computer vision and other fields.

[0061] Embedding vectors: Embedding vectors are essentially a mapping that maps objects in a high-dimensional discrete space to a low-dimensional continuous vector space. For example, in natural language processing, each word in a vocabulary is originally a discrete symbol, but through embedding, it is converted into a low-dimensional real vector. This allows the model to better capture the semantic and grammatical relationships between objects, and the embedding vectors corresponding to similar objects are closer in the vector space.

[0062] Support Vector Machine (SVM) model: A supervised machine learning model that can be used for classification and regression analysis.

[0063] Genotype: The genotype is the sum of all the genetic material within an organism's cells. This genetic material contains a specific combination of genes that determine the organism's various inherited traits. Genes are segments of DNA that have heritable effects. Different genes carry different genetic information, and the genotype is the combination of these genes.

[0064] Cerebrovascular disease (stroke) is a common neurological disorder with high rates of disability and mortality, posing a significant global public health burden. Genetic factors play a significant role in the development and progression of cerebrovascular disease. Polygenic risk scores (PRSs) assess individual genetic susceptibility by integrating information from multiple disease-associated genetic variants (typically single nucleotide polymorphisms (SNPs)). They have become an important tool for predicting risk for complex diseases. Conventional PRS models typically directly use a large number of SNPs as features. However, this approach suffers from high dimensionality, making model training difficult, interpreting biological significance difficult, and insufficient information utilization. Specifically, the large number of SNPs involved (tens of thousands or even millions) complicates model training, makes overfitting prone, and requires a large number of samples to achieve robust results. The cumulative effects of a large number of SNPs are difficult to intuitively link to biological pathways or functions, limiting our understanding of disease mechanisms. Furthermore, relying solely on the presence or absence of SNPs or genotype information overlooks the complex interactions between genes in their function and expression regulation.

[0065] Based on this, to address the inaccurate prediction issues caused by directly using SNPs as features to process genetic data in related technologies, the present invention provides a genetic data processing method, system, electronic device, and storage medium, in order to improve the accuracy of genetic data processing. The genetic data processing method provided in the present invention is described below.

[0066] Reference Figure 1 In some embodiments, the method for processing genetic data provided in the embodiments of the present application includes but is not limited to steps S101 to S104.

[0067] Step S101 : obtaining genotype data of an individual to be tested, and determining a plurality of associated genes of the individual to be tested based on the genotype data and single nucleotide polymorphism data associated with a target risk task.

[0068] Among them, the method for processing genetic data provided by the embodiment of the present disclosure can specifically be a method for predicting the risk score of a disease. In the embodiment of the present disclosure, before the risk assessment of the individual to be tested is performed, a gene feature library can be pre-constructed, and a risk score prediction model for a specific target risk task, or a prediction model, can be trained based on the pre-constructed gene feature library. Then, the trained prediction model is used to predict the risk score of the target risk task for the individual to be tested. Specifically, when predicting the risk score of the individual to be tested, the genotype information of the individual to be tested can be obtained first, and then the associated genes of the individual to be tested can be determined based on the genotype information of the individual to be tested; thereafter, the gene features corresponding to these associated genes are further found in the pre-constructed gene feature library, and the gene features corresponding to these associated genes are fused to obtain individual features; finally, the trained prediction model is used to predict the risk score of the fused individual features to obtain an accurate risk score for the individual to be tested.

[0069] Below, we will introduce in detail the three aspects of the construction process of the gene feature library, the training process of the prediction model, and the process of using the trained prediction model to perform risk scoring on the tested individuals.

[0070] like Figure 2 FIG. 1 is a flow chart of a method for constructing a gene signature library provided in this application. In some embodiments, the process of constructing a gene signature library includes:

[0071] Step S201: Acquire a plurality of first training samples, where the first training samples include gene expression profiles of human samples.

[0072] In an embodiment of the present application, a method for extracting gene embedding vectors based on a deep learning model is provided. Specifically, the deep learning model can be a model based on the Transformer architecture, such as a Transformer-based bidirectional encoder (BERT) or an autoregressive model based on a generative pretrained Transformer (GPT). In an embodiment of the present application, the deep learning model used to construct the gene signature library can be referred to as an embedding model.

[0073] After determining the model architecture of the embedding model for constructing the gene signature library, a plurality of first training samples for model training can be further obtained, wherein the plurality of first training samples can include gene expression profiles of human samples. That is, in an embodiment of the present application, a plurality of gene expression profiles can be determined based on the population to which the individual to be tested belongs, and then the trained embedding model is used to extract the gene embedding representation of the genes corresponding to the population to which the individual to be tested belongs to construct the gene signature library. Among them, the gene expression profile refers to the collection of all gene transcription products of a specific cell, tissue or organism in a certain functional state, which can reflect the gene expression situation. The gene expression process is that DNA is transcribed into RNA and then translated into protein. Gene expression spectrum analysis focuses on the transcription level, detecting the type and abundance of mRNA in the cell to understand the gene expression state. For example, when the individual to be tested is human, large-scale human gene expression profiles can be collected, such as batch RNA sequencing data from public databases such as ARCHS4, GTEx, GeneExpression Omnibus (GEO), Sequence Read Archive (SRA), etc.

[0074] Step S202: Use the first training sample to perform self-supervised learning on the embedding model.

[0075] Among them, in the embodiment of the present application, the method for training the embedded model can be self-supervised learning training. Among them, the core of self-supervised learning is to construct a pre-training task to allow the model to mine supervisory signals from the data. The model is trained on the pre-training task and learns the intrinsic structure and feature representation of the data. These representations can be used as initial features for downstream tasks to improve the performance of the model in downstream tasks. Self-supervised training does not require a large amount of labeled data, so that large-scale data training can be achieved on the basis of low data labeling costs, ensuring the training effect of the model.

[0076] During the training of the embedding model, the training objectives can be designed in one of the following two ways to obtain high-quality gene embedding representations: one way is to predict the true identity of the masked gene (such as the gene identifier) ​​based on the gene expression context information. This task aims to capture the "semantic" relationship of genes in the expression network (such as the co-regulatory relationship in the co-expression network); the other way is to predict the expression level of the masked gene at the numerical level. The expression level can be predicted by discretization (such as binning), thereby strengthening the model's learning of gene expression intensity information.

[0077] After constructing the training samples, preprocessing is performed based on the gene expression profiles of each training sample, such as normalizing the gene expression of each training sample. Normalization can include calculating a Z-score or performing sorting and binning. Then, based on the normalization results (e.g., sorting information or binning results based on the Z-score), target training samples are determined. Self-supervised learning of the embedding model is then performed based on the target training samples (including the sorted gene identifier sequences and the normalization results).

[0078] It is understood that during the self-supervised learning training of the embedding model, the target training sample can be used as the model input. The gene expression levels of some genes (or the gene identifiers of some genes) are then randomly masked, allowing the embedding model to predict the true expression levels (or true identities) of the masked genes. The training targets are generated by leveraging the intrinsic structure of the data without the need for manual labeling. After training is completed, the sample gene expression data can be further used as input for the inference phase. The unmasked gene identifier sequence obtained after preprocessing and sorting is used as the input of the embedding model, allowing the embedding model to extract the embedded representation of each gene, for example, extracting the multi-dimensional vector of the input layer (or a hidden layer) of the embedding model as the embedded representation of each gene.

[0079] Furthermore, the two self-supervised learning methods mentioned above can be trained jointly. For example, the same Transformer encoder can be used to simultaneously learn the contextual dependence and numerical distribution of gene expression, set different prediction heads (predicting expression level or true identity), and combine the losses of the two tasks (which can be weighted and used as model training parameters or dynamically adjusted according to task importance or convergence speed (such as focusing on gene ID prediction in the early stage and expression level prediction in the later stage), so that the model can predict both tasks at the same time and learn a more comprehensive embedding representation.

[0080] Step S203: When the embedding model training converges, a gene feature library is constructed based on the embedding representations of different genes extracted by the trained embedding model.

[0081] The embedding model training process is an iterative process. During this process, the number of iterations can be counted and the change in model parameters can be detected. When the number of iterations reaches a preset number, or the change in model parameters is less than a preset change threshold, the embedding model training is considered converged, resulting in a trained embedding model.

[0082] After the embedding model is trained, the embedding vector corresponding to each gene can be extracted from the trained embedding model as the embedding feature of the gene. Specifically, the embedding vector corresponding to each gene can be extracted from the input layer or a hidden layer of the model as the embedding feature of the gene. These embedding vectors can be low-dimensional dense real number vectors that encode the function and relationship information of the gene in the expression network. Specifically, the dimension of the embedding vector can be tens to hundreds of dimensions, for example, 32 dimensions, or in some embodiments, it can be reduced to a lower dimension by an automatic encoder or the like. After obtaining the embedding features corresponding to multiple genes, a gene feature library can be constructed based on the embedding features of these multiple genes.

[0083] In the examples of this application, gene embedding vectors are obtained by training with large-scale gene expression data, which can capture the functional roles and relationships of genes in complex biological networks, thereby compressing high-dimensional and sparse gene information into a low-dimensional and dense vector space. These embedding vectors contain rich biological information that the model autonomously learns from data-driven gene expression patterns, such as gene function, biological processes involved, pathway information, and even associations with diseases.

[0084] After constructing the gene signature library, one or more risk score prediction models can be further trained based on the pre-constructed gene signature library. In the present embodiment, the risk score prediction model can be referred to as a prediction model, which can specifically be a machine learning classification model, such as logistic regression, support vector machine (SVM), random forest, gradient boosting machine, etc.

[0085] like Figure 3 FIG. 1 is a flow chart of the model training method provided in this application. In an embodiment of this application, the process of training the prediction model includes:

[0086] Step S301: Acquire multiple second training samples.

[0087] In an embodiment of the present application, the prediction model can be trained in a supervised training manner, that is, the above-mentioned prediction model is supervisedly trained using labeled sample data. In this way, a labeled second training sample can be obtained first, and the labeled second training sample can include sample genotype data of multiple sample individuals and a phenotypic information label corresponding to each sample individual. For example, when the sample individual is a human individual and the target risk task is to perform a cerebrovascular disease score prediction task, the genotype data of multiple sample individuals and whether each sample individual is a cerebrovascular disease patient can be obtained. When the sample individual is a cerebrovascular disease patient, the corresponding phenotypic information label is determined to be 1; conversely, when the sample individual is not a cerebrovascular disease patient, the corresponding phenotypic information label is determined to be 0.

[0088] Step S302 : determining the sample-associated gene of each sample individual based on the sample genotype data, and searching the gene signature corresponding to each sample-associated gene in the gene signature library to obtain a plurality of sample-associated gene signatures.

[0089] When training the prediction model based on the acquired multiple second training samples, the sample-associated genes corresponding to the individual sample can be first determined based on the sample genotype data contained in each second training sample. Specifically, determining the sample-associated genes corresponding to the individual sample based on the sample genotype data can first be performed by mapping the variant sites in the sample genotype data to determine a subset of associated genes. Subsequently, the genes mapped to the single nucleotide polymorphism variant sites significantly associated with the target risk task can be used as another subset of associated genes. These two subsets of associated genes are integrated to obtain the sample-associated genes corresponding to each individual sample.

[0090] After determining the sample-associated genes corresponding to each sample individual, the gene features corresponding to each sample-associated gene can be searched in the aforementioned pre-constructed preset gene library for the multiple sample-associated genes corresponding to each sample individual, thereby obtaining the multiple sample-associated gene features corresponding to each sample individual.

[0091] Step S303: Integrate the gene features associated with multiple samples to obtain individual sample features, and input the individual sample features into the prediction model to be trained to perform risk prediction and obtain a predicted risk score.

[0092] After the multiple sample-associated gene features corresponding to each sample individual are determined, the multiple sample-associated gene features corresponding to each sample individual can be integrated to obtain the sample individual features corresponding to each sample individual.

[0093] The feature integration of genes associated with multiple samples can be performed by simple averaging of corresponding dimensions, weighted averaging, or direct vector concatenation. Alternatively, in some embodiments, a pooling-based aggregation method, a convolution-based aggregation method, an attention-based aggregation method, or a linear or nonlinear dimensionality reduction aggregation method can be used. These aggregation methods are described in detail below.

[0094] In this way, after integrating the sample individual characteristics corresponding to the sample individual, the sample individual characteristics can be input into the prediction model to be trained to estimate the risk score of the target risk task, and obtain the predicted risk score corresponding to the sample individual predicted by the prediction model.

[0095] Step S304 : Calculate the loss value according to the predicted risk score and the corresponding phenotypic information label, and update the parameters of the prediction model based on the loss value.

[0096] Furthermore, a loss value can be calculated based on the predicted risk score for the individual sample output by the prediction model and the phenotypic information label associated with the target risk task for that individual sample. Specifically, the cross-entropy method can be used to calculate the loss value. After calculating the loss value, the backpropagation gradient can be determined based on the loss value, and then the gradient backpropagation process can be performed to update the parameters of the prediction model.

[0097] The process of training the prediction model may include iteratively updating the model parameters of the prediction model. Specifically, the training samples in the second training sample may be divided into multiple batches, and then the model parameters of the prediction model may be updated in batches. When the number of update rounds reaches a preset number, or when the amplitude of the change in the model parameters is detected to be less than a preset value, it can be determined that the model training has reached a convergence state, the model training is determined to be complete, and the final model parameters are used as the model parameters of the trained prediction model.

[0098] After training the prediction model using this method, the model's effectiveness can be further evaluated. Specifically, cross-validation and other methods can be used to evaluate the effectiveness of the trained prediction model. Once the prediction model's effectiveness is determined to be satisfactory, the prediction model can be deployed online and used to predict risk scores for the target risk tasks mentioned above.

[0099] When it is necessary to use the deployed prediction model to estimate the risk score of a target risk task for a certain individual to be tested, the genotype data of the individual to be tested can be obtained first, and then multiple associated genes of the individual to be tested can be determined based on the genotype data of the individual to be tested and the single nucleotide polymorphism data associated with the target risk task.

[0100] In some embodiments, determining multiple associated genes of the individual to be tested based on genotype data and single nucleotide polymorphism data associated with the target risk task includes:

[0101] obtaining first single nucleotide polymorphism data associated with a target risk task;

[0102] Based on the statistical test results of the allele frequency difference between the risk individual and the reference individual of the target risk task, the first single nucleotide polymorphism data are screened to obtain a plurality of second single nucleotide polymorphism data;

[0103] A plurality of third single nucleotide polymorphism data are determined based on the genotype data, and a plurality of associated genes of the individual to be tested are determined according to the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data.

[0104] And, in some embodiments, determining a plurality of associated genes of the individual to be tested based on the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data includes:

[0105] Obtaining genomic locations and functional association information corresponding to the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data;

[0106] The plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data are mapped to corresponding genes according to the genomic position and the functional association information to obtain a plurality of associated genes.

[0107] In an embodiment of the present disclosure, a method for obtaining a complete set of associated genes from two dimensions is provided. Specifically, SNPs significantly associated with the target risk task can be first obtained based on GWAS (i.e., first single nucleotide polymorphism data). For example, if the target risk task is cerebrovascular disease risk prediction, SNP information related to cerebrovascular disease can be obtained from the latest GWAS studies. Alternatively, all SNP information located in specific gene regions (such as coding regions or promoters) can be obtained. Then, statistical tests can be performed using the allele frequency differences between individuals with the target risk task and healthy individuals (reference individuals) to screen for significantly associated SNPs. The statistical tests can be specifically performed using the chi-square test or Fisher's exact test. When screening SNPs, significantly associated SNPs with a p-value less than 0.05 (i.e., second single nucleotide polymorphism data) can be selected. Furthermore, bioinformatics tools such as SnpEff and VEP can be used to map these SNPs to corresponding genes based on their genomic location and functional association information, ensuring consistency in the mapping of the same SNP to genes across individuals. In this way, the associated genes of the first dimension can be obtained.

[0108] Furthermore, the SNP sites of the genotype (i.e., third single nucleotide polymorphism data) can be determined based on the individual's genotype data. Bioinformatics tools such as SnpEff and VEP can then be used to map the SNP variant sites corresponding to the genotype to relevant genes, taking into account the physical location of the SNP (e.g., within the gene, in the promoter region, in adjacent regions) and functional association information (e.g., eQTL data). This allows the second dimension of associated genes for the individual being tested to be obtained. Ultimately, the associated genes of these two dimensions are combined to obtain the associated genes for the individual being tested. This standardized mapping process ensures that the method for determining associated genes remains consistent during the training and inference of the risk score prediction model, thereby improving the accuracy and reliability of risk score predictions.

[0109] Step S102: query the pre-built gene feature library for the associated gene feature corresponding to each associated gene.

[0110] After determining the multiple associated genes corresponding to the individual to be tested using the above method, the gene signature of each associated gene can be further queried based on the pre-established gene signature library described in step S101, thereby obtaining the gene signature corresponding to each associated gene. In other words, multiple associated gene signatures of the individual to be tested can be obtained by querying the gene signature library based on the multiple associated genes.

[0111] Specifically, the gene names and corresponding gene features in the gene feature library can be stored in the form of key-value pairs. Then, when querying, the key data corresponding to the associated genes of the individual to be tested can be entered to obtain the corresponding value data returned by the query.

[0112] Step S103: integrating the features of multiple associated genes to obtain individual features.

[0113] After obtaining multiple associated gene features corresponding to multiple associated genes of the individual to be tested, these associated gene features can be further integrated to generate a single feature vector representing the individual to be tested and related to the target risk task.

[0114] In some embodiments, multiple associated gene signatures are integrated to obtain individual signatures, including:

[0115] Obtain weight coefficients corresponding to multiple associated gene features;

[0116] Multiple associated gene features are weighted based on the weight coefficient to obtain individual features.

[0117] In an embodiment of the present application, a feature integration method is provided for obtaining individual features by weighted calculation of multiple associated gene features based on associated gene feature weight coefficients. Among them, for different associated genes, the degree of association with the target risk task is different, and different weights can be set for different associated gene features according to the difference in the degree of association. When the correlation between the associated gene and the target risk task is strong, a higher weight is set for the associated gene. Conversely, when the correlation between the associated gene and the target risk task is weak, a lower weight can be set for the associated gene. In this way, the correlation between the individual feature and the target risk task can be further improved, thereby improving the accuracy of the disease risk score prediction for the target risk task.

[0118] In some embodiments, multiple associated gene signatures are integrated to obtain individual signatures, including:

[0119] Perform convolution processing on the feature sequence composed of multiple related gene features to obtain sequence features;

[0120] Based on sequence features, multiple associated gene features are integrated to obtain individual features.

[0121] In an embodiment of the present application, a feature integration method based on convolution operations is provided. Specifically, a feature sequence composed of multiple correlated gene features can be convolved to obtain a sequence feature. The sequence feature can include local correlations and dynamic patterns between adjacent genes. Based on the sequence feature, the features of multiple correlated genes can then be integrated to obtain an individual feature.

[0122] In some embodiments, feature integration of multiple associated gene features may also employ a pooling-based aggregation operation. Specifically, methods such as maximum pooling or mean pooling may be employed to sample and aggregate multiple associated gene features, thereby extracting the most representative local information of the genetic characteristics of the individual being tested.

[0123] In some embodiments, feature integration of multiple associated gene features may also be performed using a feature integration method based on an attention mechanism. Specifically, weights may be adaptively assigned to different associated gene features based on the attention mechanism, thereby highlighting gene information that is more critical to the prediction results.

[0124] In some embodiments, feature integration of multiple associated gene features can also employ feature integration methods based on linear or nonlinear dimensionality reduction. Specifically, dimensionality reduction methods such as principal component analysis, linear transformation, and autoencoders can be combined to compress the high-dimensional feature vectors aggregated using these methods, thereby extracting low-dimensional features that best reflect the individual's genetic information.

[0125] The above-mentioned complex aggregation method can fully explore the intrinsic connections between different genetic features, improve the discriminative ability of individual feature vectors, and provide more biologically meaningful feature support for subsequent polygenic risk scoring models for target risk tasks based on genetic features.

[0126] It is understood that the above embodiments provide a variety of feature integration methods, and in order to ensure the accuracy of the genetic risk score of the target risk task, it is necessary to ensure that a consistent feature integration method is used when training and inferring the risk score prediction model corresponding to the target risk task. For example, when training the risk score prediction model corresponding to the target risk task, a feature integration method based on convolution operation is used to integrate multiple sample-related features of the sample individual; then, when using the trained risk score prediction model corresponding to the target risk task to predict the risk score of the target risk task for the individual to be tested, it is also necessary to use a feature integration method based on convolution operation to integrate the multiple associated gene features of the individual to be tested, thereby ensuring the accuracy of the predicted risk score prediction of the target risk task.

[0127] Step S104: calling the prediction model to process the individual characteristics to obtain the risk score corresponding to the individual to be tested and the target risk task.

[0128] After the individual characteristics of the individual to be tested are integrated, the trained prediction model can be called to predict the risk score of the target risk task based on the individual characteristics, thereby obtaining the risk score corresponding to the individual to be tested and the target risk task.

[0129] The genetic data processing method provided in this application greatly reduces the complexity of the risk score prediction model and the sample size requirement by replacing high-dimensional sparse (tens of thousands) SNP features with low-dimensional dense (e.g., dozens to hundreds of dimensions) gene embedding vectors, thereby improving the training efficiency and generalization ability of the model.

[0130] Furthermore, gene signatures (i.e., gene embedding vectors) are learned from large-scale gene expression data and contain rich information about gene functions, interactions, and pathways. Using these embeddings as features allows this biological context to be incorporated into the PRS model, helping it capture deeper genetic risk factors. Gene embeddings capture the complex relationships between genes and better reflect true biological effects than simple SNP summation. Therefore, PRS models based on gene embedding features can significantly improve the accuracy of risk score predictions.

[0131] Although the embedding vector itself is abstract, we can explore potential pathogenic pathways or mechanisms by analyzing the genes and their functions corresponding to the embedding vectors that contribute greatly to the prediction results, providing clues for subsequent functional studies.

[0132] The following is a detailed description of the risk score prediction method provided by this application using a specific example. This example specifically provides a risk score prediction method for human cerebrovascular disease.

[0133] First, construct a gene embedding vector library.

[0134] Data source selection: Approximately 500,000 human bulk RNA-seq sample data from the ARCHS4 database were used.

[0135] Data preprocessing: Normalize the gene expression data for each sample, for example, by calculating the Z-score of each gene in each sample. Then, within each training sample, sort the genes in descending order based on their Z-scores, and select a certain number of genes that rank high as target training samples.

[0136] Model training: A 6-layer Transformer model based on the BERT architecture was selected. The input was the gene identifiers and their ranking information corresponding to the sample data. The training task was to randomly mask 15% of the gene identifiers in the input data and predict the original gene identifiers of the masked genes based on the unmasked genes and their expression ranking information. The AdamW optimizer was used for model training.

[0137] Embedding vector extraction: After training convergence, extract the 32-dimensional embedding vector corresponding to each gene in the model input layer (or a hidden layer) to construct a gene embedding vector library.

[0138] Then, a risk score prediction model for human cerebrovascular disease was trained.

[0139] Sample Data Preparation: A synthetic dataset of SNPs significantly associated with cerebrovascular disease (CVD) was used, using a 1KG healthy population panel and GWAS analysis. Relevant information from the GWAS summary data (basic variant information, OR values, allele frequencies, etc.) was integrated to select SNPs with confirmed significant associations with CVD (80 SNPs in total). Genotype information (phase 1, v3) for the corresponding SNPs in the 1KG healthy population was obtained to construct a training sample. Case samples were simulated based on the genotype distribution of the SNPs in the GIGASTROKE cohort. (For example, if the bi-allelic distribution of a SNP in the cohort is AA 50%, AC 30%, and CC 20%, the simulated case samples were assigned the phase information of the SNP using this information, i.e., 50% probability of 0|0, 30% probability of 1|0 or 0|1, and 20% probability of 1|1).

[0140] Genotype-gene association: Using public databases (e.g., dbSNP, GENCODE) and tools (e.g., VEP, ANNOVAR), each individual's SNPs and genetic variants are annotated to their corresponding or adjacent genes. Furthermore, brain tissue eQTL data from databases such as GTEx can be used to associate SNPs significantly associated with gene expression with their corresponding genes.

[0141] Generation of sample individual characteristics: For each sample individual, first identify all SNPs in their genome that are associated with known cerebrovascular disease risks or all SNPs located in specific gene regions (such as coding regions, promoters). Then, based on the association between genotype and gene, determine the gene set corresponding to these SNPs. The genes associated with individual i are denoted as {Gene_A, Gene_B, Gene_C, ...}. Find the embedding vectors corresponding to Gene_A, Gene_B, Gene_C... from the gene embedding vector library, .... Then calculate the weighted average of these embedding vectors as the feature vector of individual i. The weight can be based on the genotype of the SNP (for example, the homozygous risk allele is 2, the heterozygous is 1, and the homozygous non-risk is 0) and / or the effect value of the SNP in the GWAS study (beta coefficient or ln(OR) value). The formula is as follows:

[0142] .

[0143] Model training: The training samples were split into a training set and a test set at an 80% / 20% ratio. A support vector machine (SVM) model was then trained using the individual feature vectors of the training set and the corresponding cerebrovascular disease labels (1 for patients and 0 for controls).

[0144] Model testing and evaluation: For each individual in the test set, their individual feature vectors are generated and then input into the trained SVM model to obtain the predicted cerebrovascular disease risk probability. Performance indicators are then calculated based on the predicted probability and the labels of the test set. Performance indicators include AUC (area under the ROC curve), confusion matrix, and F1 score. Figure 4 As shown in FIG, the area under the ROC curve of the risk score prediction model provided by this application in the independent test set of simulated data is shown. Figure 5 As shown in FIG, the F1 score diagram of the risk score prediction model provided by this application in the independent test set of simulated data. Figure 6 As shown in FIG, it is a schematic diagram of the confusion matrix of the risk score prediction model provided by this application in the independent test set of simulated data. Figure 7Figure 2 shows a schematic diagram evaluating the importance of features across gene vector dimensions. Overall, after five-fold cross-validation, the optimal hyperparameter model (kernel: linear, gamma: 1.5, C: 10) achieved good classification performance (AUC=0.87, Max F1Score=0.85). The confusion matrix (c) indicates the model's potential false positive risk, urging caution when interpreting positive results. The feature contribution (d), calculated by randomly shuffling each feature value and then calculating the degree of change in the model's classification performance on the test set (boxplots show the distribution of the results of randomly shuffling feature values), indicates that features such as the 16th dimension of the gene vector (Feature_17, starting from 0) are important for classifying cerebrovascular disease risk. Further dimensionality reduction is considered possible to reduce computational costs and improve model performance.

[0145] Finally, the cerebrovascular disease risk score prediction model was used to predict the cerebrovascular disease risk score.

[0146] The trained cerebrovascular disease risk scoring model can be deployed online. When a cerebrovascular disease risk score prediction is needed for an individual, the individual's genotype data can be obtained first. Based on the genotype data, the individual's associated genes can be determined. Based on the associated genes, the corresponding gene embedding vectors are searched in the gene embedding vector library and integrated into an individual feature vector. Finally, the trained cerebrovascular disease risk scoring model is called to process the individual feature vector and output the individual's cerebrovascular disease risk score.

[0147] like Figure 8a and 8b The figure below shows a comparison of the ROC and F1 scores of the SNP-based scoring model in the solution provided in this application and related technologies in an independent test set. Both demonstrate that the present model has stronger classification performance and robustness. However, under certain threshold conditions (approximately <0.1) in the related technology method, the true positive detection rate is 0, so the F1 value does not exist.

[0148] In some embodiments, the present application may also provide a cerebrovascular disease risk prediction system based on cloud computing.

[0149] Backend implementation:

[0150] Precompute gene embeddings using Python and a deep learning framework (such as TensorFlow or PyTorch) (this can be done on a high-performance computing cluster). Store the resulting gene embedding vectors in an efficient key-value database (such as Redis) or vector database. Implement the logic for the feature generation module, model training module, and risk prediction module using Python and a machine learning library (such as Scikit-learn and XGBoost). Use a relational database (such as PostgreSQL) to store user information, genotype data, phenotypic data, and model parameters. Provide a service interface via a RESTful API.

[0151] Front-end implementation:

[0152] Develop a web interface that allows authorized users to upload individual genotype data files (e.g., VCF format). The front-end calls the back-end API and submits the data for analysis. The back-end performs feature generation and risk prediction, returning the results (risk score, probability, confidence interval, etc.) to the front-end. The front-end displays the prediction results in charts or reports.

[0153] System workflow:

[0154] The user logs in to the system and uploads the genotype file of the person to be tested. The system then receives the file and calls the feature generation module (queries the gene embedding library and calculates the individual feature vector). It then calls the risk prediction module (loads the trained model and makes predictions). The prediction results are then returned to the user interface, and the user finally views the risk report.

[0155] Reference Figure 9 In some embodiments, the present application also provides a genetic data processing system 900, which includes:

[0156] The data acquisition unit 910 is used to obtain genotype data of the individual to be tested, and determine multiple associated genes of the individual to be tested based on the genotype data and single nucleotide polymorphism data associated with the target risk task;

[0157] A query unit 920 is configured to query a pre-constructed gene signature library for associated gene signatures corresponding to each associated gene, wherein the gene signature library includes embedded representations of different genes constructed based on an embedding model; the embedding model is obtained by self-supervised learning of multiple gene expression profiles of the same species as the individual to be tested;

[0158] An integration unit 930 is used to integrate multiple related gene features to obtain individual features;

[0159] The prediction unit 940 is used to call the prediction model to process individual characteristics to obtain a risk score corresponding to the individual to be tested and the target risk task. The prediction model is trained based on sample characteristics and corresponding phenotypic information of multiple sample individuals.

[0160] In some embodiments, the acquisition unit includes:

[0161] an acquisition subunit, configured to acquire first single nucleotide polymorphism data associated with a target risk task;

[0162] a screening subunit, configured to screen the first single nucleotide polymorphism data to obtain a plurality of second single nucleotide polymorphism data based on a statistical test result of an allele frequency difference between a risk individual of a target risk task and a reference individual;

[0163] The determination subunit is used to determine a plurality of third single nucleotide polymorphism data based on the genotype data, and to determine a plurality of associated genes of the individual to be tested according to the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data.

[0164] Optionally, in some embodiments, determining the subunit includes:

[0165] An acquisition module, configured to acquire genome locations and functional association information corresponding to a plurality of second single nucleotide polymorphism data and a plurality of third single nucleotide polymorphism data;

[0166] The mapping module is used to map the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data to corresponding genes according to the genome location and the functional association information to obtain a plurality of associated genes.

[0167] Optionally, in some embodiments, the present application further provides a gene signature library construction device, comprising:

[0168] an acquisition unit, acquiring a plurality of first training samples, wherein the first training samples include gene expression profiles of human samples;

[0169] A training unit, which uses the first training sample to perform self-supervised learning on the embedding model;

[0170] The construction unit is used to construct a gene feature library based on the embedding representations of different genes extracted by the trained embedding model when the embedding model training converges.

[0171] Optionally, in some embodiments, the training unit includes:

[0172] a normalization subunit, configured to perform normalization processing on the gene expression profile of each first training sample, and determine a target training sample based on the normalization processing result;

[0173] The training subunit is used to obtain an embedding model through self-supervised learning using random masks based on target training samples.

[0174] Optionally, in some embodiments, the integration unit includes:

[0175] The second acquisition subunit is used to obtain weight coefficients corresponding to multiple associated gene features;

[0176] The calculation subunit is used to perform weighted calculation on multiple associated gene features based on the weight coefficient to obtain individual features.

[0177] Optionally, in some embodiments, the integration unit includes:

[0178] The convolution subunit is used to perform convolution processing on the feature sequence composed of multiple related gene features to obtain sequence features;

[0179] The integrator unit is used to integrate multiple related gene features based on sequence features to obtain individual features.

[0180] Optionally, in some embodiments, the present application further provides a model training device, comprising:

[0181] A third acquisition subunit is used to acquire a plurality of second training samples, where the second training samples include sample genotype data of a plurality of sample individuals and corresponding phenotypic information labels;

[0182] A search subunit is used to determine the sample-associated gene of each sample individual based on the sample genotype data, and to search the gene signature corresponding to each sample-associated gene in the gene signature library to obtain multiple sample-associated gene signatures;

[0183] The prediction subunit is used to integrate the gene features associated with multiple samples to obtain individual sample features, and input the individual sample features into the prediction model to be trained to predict the disease risk and obtain a predicted risk score;

[0184] The updating subunit is used to calculate the loss value according to the predicted risk score and the corresponding phenotypic information label, and update the parameters of the prediction model based on the loss value.

[0185] Reference Figure 10 , Figure 10 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:

[0186] The processor 1001 can be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0187] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called by the processor 1001 to execute the genetic data processing method of the embodiments of this application.

[0188] Input / output interface 1003, used to implement information input and output;

[0189] Communication interface 1004, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0190] Bus 1005 , which transmits information between various components of the device (e.g., processor 1001 , memory 1002 , input / output interface 1003 , and communication interface 1004 );

[0191] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0192] The present application also provides a computer program product, which includes a computer program. A processor of a computer device reads and executes the computer program, so that the computer device executes the above-mentioned genetic data processing method.

[0193] The terms "first," "second," "third," "fourth," and the like (if any) in the specification of the present disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present disclosure described herein, for example, can be implemented in orders other than those illustrated or described herein. In addition, the terms "comprises" and "comprising," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0194] It should be understood that in the present disclosure, "at least one (item)" refers to one or more, and "plurality" refers to two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0195] It should be understood that in the description of the embodiments of the present application, the meaning of multiple (or multiple items) is more than two, greater than, less than, exceed, etc. are understood to exclude the number itself, and above, below, within, etc. are understood to include the number itself.

[0196] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0197] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0198] In addition, the functional units in the various embodiments of the present disclosure may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0199] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the various embodiments of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0200] It should also be understood that the various implementation methods provided in the embodiments of the present application can be combined arbitrarily to achieve different technical effects.

[0201] The above is a specific description of the implementation methods of the present disclosure, but the present disclosure is not limited to the above implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present disclosure. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present disclosure.

Claims

1. A method for processing genetic data, characterized in that: The method comprises: Obtaining genotype data of the individual to be tested, and obtaining first single nucleotide polymorphism data associated with the target risk task; Based on the statistical test results of the allele frequency difference between the risk individual and the reference individual of the target risk task, the first single nucleotide polymorphism data are screened to obtain a plurality of second single nucleotide polymorphism data; Determining a plurality of third single nucleotide polymorphism data based on the genotype data, and obtaining genomic positions and functional association information corresponding to the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data; Mapping the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data to corresponding genes according to the genomic positions and the functional association information to obtain a plurality of associated genes; querying a pre-constructed gene signature library for associated gene signatures corresponding to each associated gene, wherein the gene signature library includes embedded representations of different genes constructed based on an embedding model; the embedding model is obtained by self-supervised learning of multiple gene expression profiles of the same species as the individual to be tested; Performing feature integration on the multiple associated gene features to obtain individual features; The prediction model is called to process the individual characteristics to obtain a risk score corresponding to the individual to be tested and the target risk task, wherein the prediction model is trained based on the sample characteristics and corresponding phenotypic information of multiple sample individuals.

2. The method according to claim 1, characterized in that The process of pre-building a gene signature library includes: Acquire a plurality of first training samples, wherein the first training samples include gene expression profiles of human samples; Performing self-supervised learning on the embedding model using the first training sample; When the embedding model training converges, a gene feature library is constructed based on the embedding representations of different genes extracted by the trained embedding model.

3. The method according to claim 2, characterized in that The self-supervised learning of the embedding model using the first training sample includes: performing normalization processing on the gene expression profile of each of the first training samples, and determining a target training sample based on the normalization processing result; Based on the target training sample, the embedding model is obtained by self-supervised learning using random masks.

4. The method according to claim 1, wherein The step of integrating the features of the multiple associated genes to obtain individual features includes: Obtaining weight coefficients corresponding to the multiple associated gene features; The plurality of associated gene features are weightedly calculated based on the weight coefficients to obtain individual features.

5. The method according to claim 1, wherein The step of integrating the features of the multiple associated genes to obtain individual features includes: Performing convolution processing on the feature sequence composed of the multiple associated gene features to obtain sequence features; The multiple associated gene features are integrated based on the sequence features to obtain individual features.

6. The method according to claim 1, characterized in that The training process of the prediction model includes: Acquire a plurality of second training samples, wherein the second training samples include sample genotype data of a plurality of sample individuals and corresponding phenotypic information labels; Determining a sample-associated gene for each sample individual based on the sample genotype data, and searching the gene signature corresponding to each sample-associated gene in the gene signature library to obtain a plurality of sample-associated gene signatures; Integrating the multiple sample-associated gene features to obtain sample individual features, and inputting the sample individual features into the prediction model to be trained to perform risk prediction to obtain a predicted risk score; A loss value is calculated according to the predicted risk score and the corresponding phenotypic information label, and the parameters of the prediction model are updated based on the loss value.

7. A genetic data processing system, characterized in that: The genetic data processing system includes: A data acquisition unit is used to obtain genotype data of an individual to be tested and obtain first single nucleotide polymorphism data associated with a target risk task; based on a statistical test result of the allele frequency difference between the risk individual of the target risk task and a reference individual, screen the first single nucleotide polymorphism data to obtain a plurality of second single nucleotide polymorphism data; based on the genotype data, determine a plurality of third single nucleotide polymorphism data, obtain genomic locations and functional association information corresponding to the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data; map the plurality of second single nucleotide polymorphism data and the plurality of third single nucleotide polymorphism data to corresponding genes according to the genomic locations and the functional association information to obtain a plurality of associated genes; a query unit, configured to query a pre-constructed gene signature library for associated gene signatures corresponding to each associated gene, wherein the gene signature library includes embedded representations of different genes constructed based on an embedding model; the embedding model is obtained by self-supervised learning of multiple gene expression profiles of the same species as the individual to be tested; an integration unit, configured to integrate the plurality of associated gene features to obtain individual features; The prediction unit is used to call the prediction model to process the individual characteristics to obtain the risk score corresponding to the individual to be tested and the target risk task, and the prediction model is trained based on the sample characteristics and corresponding phenotypic information of multiple sample individuals.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method for processing genetic data according to any one of claims 1 to 6 is implemented.

9. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for processing genetic data according to any one of claims 1 to 6 is implemented.

Images