Breeding trait resolution method, apparatus, device, medium, and computer program product

Abstract

The present application relates to the field of breeding trait analysis, and provides a breeding trait analysis method, device, equipment, medium and computer program product.The method comprises the following steps: constructing a target data set based on genotype data and phenotype trait data of a breeding sample; training a trait prediction model through the target data set; analyzing the effect of a genetic variation site on a phenotype trait through the trait prediction model to obtain an effect matrix; the genetic variation site is determined based on the genotype data; and performing aggregation processing on the effect matrix to obtain a key variation site corresponding to a target trait; the target trait is selected from the phenotype trait data according to requirements.The present application mines the effect of a genetic site on a trait through a trait prediction model, fully utilizes genotype data and phenotype data, and improves the accuracy of breeding trait analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of breeding trait analysis, and in particular to a breeding trait analysis method, device, equipment, medium and computer program product. Background Art

[0002] Existing methods for analyzing breeding traits rely on association analysis between genetic data and traits, requiring a large number of samples to detect significant associations. The analysis results are mainly based on additive effects (the sum of the independent contributions of multiple gene loci to a phenotype, with each allele affecting the phenotype in a linear manner). There is a lack of exploration of dominant effects (the phenomenon in which the dominant-recessive relationship between the same gene loci causes the phenotype to deviate from the additive model) and epistatic effects (the interaction between different gene loci causes the phenotypic expression of a gene to depend on the state of other genes), resulting in inaccurate analysis of crop breeding traits. Summary of the Invention

[0003] The present invention provides a breeding trait analysis method, device, equipment, medium and computer program product, which are used to solve the defect of inaccurate analysis in existing breeding trait analysis schemes and achieve accurate analysis of breeding traits.

[0004] The present invention provides a breeding trait analysis method, comprising the following steps:

[0005] Construct the target dataset based on the genotype data and phenotypic trait data of the breeding samples;

[0006] Obtaining a trait prediction model through training of the target data set;

[0007] Analyzing the effect of gene variation sites on phenotypic traits using the trait prediction model to obtain an effect matrix; the gene variation sites are determined based on the genotype data;

[0008] The effect matrix is ​​aggregated to obtain key variation sites corresponding to the target trait; the target trait is selected from the phenotypic trait data according to needs.

[0009] According to a breeding trait analysis method provided by the present invention, constructing a target data set based on genotype data and phenotypic trait data of breeding samples comprises:

[0010] The genotype data of the breeding samples are transformed to obtain the genotype matrix;

[0011] Encode the phenotypic trait data of the breeding samples to obtain the trait coding matrix;

[0012] Based on the numbering of the breeding samples, the genotype matrix and the trait coding matrix are integrated to obtain a target data set.

[0013] According to a breeding trait analysis method provided by the present invention, the trait prediction model includes an encoding module, a sampling module, a decoding module and a prediction module; the trait prediction model obtained by training the target data set includes:

[0014] The genotype matrix is ​​used as a model input, and the compressed expression matrix of the genotype data is calculated by the encoding module;

[0015] Calculating and generating an expression matrix through the sampling module and the compressed expression matrix;

[0016] The reconstructed expression matrix is ​​calculated by the decoding module and the generated expression matrix; the gene reconstruction loss is determined based on the genotype matrix and the reconstructed expression matrix;

[0017] Calculating predicted phenotypic data of the genotype matrix by the prediction module; determining phenotypic prediction loss based on the phenotypic trait data and the predicted phenotypic data;

[0018] The trait prediction model is back-propagated based on a loss function to update the network parameters of the trait prediction model; the loss function is determined based on the gene reconstruction loss and the phenotype prediction loss.

[0019] According to a breeding trait analysis method provided by the present invention, the genotype data of the breeding sample is converted to obtain a genotype matrix, which includes:

[0020] Determining gene locus data of each sample based on the genotype data of the breeding samples; the gene locus data includes identical homozygous sites, different homozygous sites, and heterozygous sites;

[0021] The gene site data is converted into numerical data to obtain a genotype matrix, wherein the same homozygous sites, the different homozygous sites and the heterozygous sites are converted into different numerical values.

[0022] According to a breeding trait analysis method provided by the present invention, the effect of gene variation sites on phenotypic traits is analyzed by the trait prediction model to obtain an effect matrix, which includes:

[0023] Determine target variant sites, total variant site set, and variant site subset based on gene variant sites;

[0024] Determining a first prediction result corresponding to the subset of variant sites using the trait prediction model;

[0025] Determining a second prediction result corresponding to a target union using the trait prediction model; the target union is determined based on the target variant site and the variant site subset;

[0026] An effect matrix of the target variant site is determined based on the total set of variant sites, the subset of variant sites, the first prediction result, and the second prediction result.

[0027] According to a breeding trait analysis method provided by the present invention, the aggregation processing of the effect matrix to obtain the key variation sites corresponding to the target trait includes:

[0028] Slice and extract the effect matrix according to the target trait to obtain a target matrix;

[0029] Performing sample aggregation processing on the target matrix to obtain a first matrix;

[0030] Performing phenotypic aggregation processing on the first matrix to obtain a second matrix;

[0031] The gene variation sites are screened based on the second matrix to obtain key variation sites corresponding to the target trait.

[0032] The present invention also provides a breeding trait analysis device, comprising the following modules:

[0033] A target dataset construction module is used to construct a target dataset based on the genotype data and phenotypic trait data of breeding samples;

[0034] A model training module, used to obtain a trait prediction model through training of the target data set;

[0035] An effect matrix construction module, configured to analyze the effects of gene variant sites on phenotypic traits using the trait prediction model to obtain an effect matrix; the gene variant sites are determined based on the genotype data;

[0036] The trait analysis module is used to aggregate the effect matrix to obtain key variation sites corresponding to the target trait; the target trait is selected from the phenotypic trait data according to needs.

[0037] The present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, any of the breeding trait analysis methods described above is implemented.

[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the breeding trait analysis methods described above.

[0039] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned breeding trait analysis methods.

[0040] The breeding trait analysis method, apparatus, equipment, medium, and computer program product provided by the present invention use genotypic and phenotypic trait data from breeding samples to construct a target dataset for training a trait prediction model. The trait prediction model then analyzes the effects of gene variant sites on phenotypic traits, mining the effects of gene loci on traits. Finally, these effects are aggregated to obtain key variant sites corresponding to the target trait. By mining the effects of gene loci on traits through the trait prediction model, this application fully utilizes genotypic and phenotypic data, thereby improving the accuracy of breeding trait analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 This is one of the flow charts of the breeding trait analysis method provided by the present invention.

[0043] Figure 2 This is the second flow chart of the breeding trait analysis method provided by the present invention.

[0044] Figure 3 It is a schematic structural diagram of the breeding trait analysis device provided by the present invention.

[0045] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0046] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0047] The following combination Figures 1-4 The present invention describes the breeding trait analysis method, device, equipment, medium and computer program product.

[0048] Figure 1This is one of the flow charts of the breeding trait analysis method provided by the present invention, such as Figure 1 As shown, the method includes the following:

[0049] Step 100: construct a target data set based on the genotype data and phenotypic trait data of the breeding samples;

[0050] First, genotype and phenotypic data for the breeding materials are collected and preprocessed, using crop numbers as units, to construct a "genotype-phenotype" dataset, the target dataset in this example. The specific process for constructing the target dataset is described in detail in subsequent examples.

[0051] Step 200: obtaining a trait prediction model through training of the target data set;

[0052] Specifically, the breeding trait analysis method provided by the present invention can use a multi-task variational autoencoder to establish a trait prediction model based on the target dataset. The multi-task variational autoencoder includes four components: an encoding module, a sampling module, a decoding module, and a prediction module.

[0053] The functional structure of each component is as follows:

[0054] The encoding module consists of M encoding components, each of which consists of a linear layer, a normalization layer, an activation layer, and a dropout layer. Each encoding component can convert an a-dimensional vector into a b-dimensional vector. The input of the first encoding component is the genotype data in the target dataset, with a dimension of The output dimension of the mth encoding component is the input dimension of the m+1th encoding component, i.e., L(m). This embodiment can use two encoding components with an output dimension of [2048, 512]. The normalization layer can use batch normalization, and the activation layer can use LeakyReLU (an improved ReLU activation function). The essence of the dropout layer is to achieve regularization by randomly masking neurons to prevent overfitting of the neural network.

[0055] The sampling module consists of a generator and two sampling components. The sampling components are linear layers, denoted as and , the input dimension is L(M), and the output dimension is the preset generation dimension , the generation dimension in this embodiment It can be 32, and the generator can use a Gaussian generator with a mean of 0 and a variance of 1.

[0056] The decoding module consists of M decoding components and 1 output layer. Each decoding component consists of a linear layer, a normalization layer, an activation layer, and a dropout layer. Each decoding component is a mirror image component corresponding to the encoding component, which can convert a b-dimensional vector into an a-dimensional vector. The input of the first decoding component is the generated dimension ; The output dimension of the mth decoding component is the input dimension of the m+1th decoding component, that is, L(M-m+1); the input dimension of the output layer is the output dimension of the Mth decoding component, that is, L(1), and the output dimension is the genotype data dimension In this embodiment, two decoding components can be used, and the output dimension is [512, 2048]. The normalization layer can use batch normalization; the activation layer can use LeakyReLU.

[0057] The prediction module includes Y prediction components, where Y is the dimension of the phenotypic trait data in the target dataset. Each prediction component includes a linear layer, a normalization layer, an activation layer, and an output layer. Each prediction component can The generating vector of dimension is enlarged to After conversion to dimensional output. For quantitative traits, the corresponding =1, when the phenotypic trait For categorical traits, is the number of categories, which corresponds to the dimension of the one-hot encoding. In this embodiment, the normalization layer can use batch normalization; the activation layer can use LeakyReLU; the magnification dimension is 128.

[0058] Step 300: Analyze the effect of gene variant sites on phenotypic traits using the trait prediction model to obtain an effect matrix; the gene variant sites are determined based on the genotype data;

[0059] Gene variant sites are screened out from the genotype data of the target data set. The present invention provides a Shapley method (a mathematical method for fairly distributing cooperative benefits in game theory) to calculate the effect value of each gene variant site on each phenotypic trait in the trait prediction model.

[0060] Step 400: Aggregate the effect matrix to obtain key variation sites corresponding to target traits; the target traits are selected from the phenotypic trait data according to needs.

[0061] From the phenotypic trait data in the target dataset, select one or more phenotypic traits as target traits. First, slice and extract the effect matrix based on the target trait to obtain the target effect matrix. Then, aggregate the target effect matrix by sample to obtain the first matrix. Aggregate the first matrix by phenotypic trait again to obtain the second matrix. Sort the second matrix in descending order, and select the gene variants with the highest effect values ​​after descending order as the key variants corresponding to the target trait.

[0062] This example uses genotypic and phenotypic data from breeding samples to construct a target dataset for training a trait prediction model. The trait prediction model then analyzes the effects of gene variant sites on phenotypic traits, mining the effects of gene loci on traits. Finally, the effects are aggregated to obtain key variant sites corresponding to the target trait. This application utilizes genotypic and phenotypic data to improve the accuracy of breeding trait analysis by mining the effects of gene loci on traits through a trait prediction model.

[0063] Figure 2 This is the second flow chart of the breeding trait analysis method provided by the present invention, such as Figure 2 As shown, the method may further include:

[0064] Step 110: converting the genotype data of the breeding sample to obtain a genotype matrix;

[0065] Step 120: Encoding the phenotypic trait data of the breeding sample to obtain a trait coding matrix;

[0066] Step 130: Based on the numbering of the breeding samples, the genotype matrix and the trait coding matrix are integrated to obtain a target data set.

[0067] Specifically, the breeding samples are numbered, and then the genotypic and phenotypic trait data are preprocessed. Genotypic data preprocessing includes quality control, gene comparison, variant detection, and conversion processing; phenotypic trait data preprocessing includes missing value filling, normalization, and encoding.

[0068] Phenotypic trait data includes two types: quantitative traits and categorical traits. Quantitative traits were subjected to missing value filling and normalization, while categorical traits were subjected to missing value filling and encoding. Table 1 shows the phenotypic trait data before preprocessing.

[0069] Table 1

[0070]

[0071] For filling missing values ​​of quantitative traits, the trait average value can be used for filling. For all quantitative traits after filling, the Min-Max normalization method can be used to put the data of quantitative traits into the interval [0, 1]. For filling missing values ​​of categorical traits, the type with the most traits can be used for filling. Each categorical trait after filling can be converted into a one-hot encoding matrix using one-hot encoding. The number of columns of the one-hot encoding matrix is ​​equal to the number of categories of the trait, and the number of rows is equal to the number of samples in the dataset. In the one-hot encoding vector of each sample, the column position corresponding to the classification value is 1, and the rest are 0. Take the quantitative traits "plant height" and "yield per mu" and the categorical trait "resistance to large leaf spot" as examples for illustration. The phenotypic trait data after preprocessing is shown in Table 2.

[0072] Table 2

[0073]

[0074] Using the sample number as an anchor, the pre-processed genotype data and phenotypic trait data are integrated to construct a "genotype-phenotype" data set, which is the target data set in this embodiment.

[0075] This embodiment processes the genotype data and phenotypic trait data of the collected breeding samples to construct a data set that can be used in a trait prediction model.

[0076] In one embodiment, the breeding trait analysis method provided by the embodiment of the present invention may further include:

[0077] Step 210: using the genotype matrix as a model input, and calculating a compressed expression matrix of the genotype data through the encoding module;

[0078] Step 220: Calculate and generate an expression matrix using the sampling module and the compressed expression matrix;

[0079] Step 230: Calculate and reconstruct an expression matrix using the decoding module and the generated expression matrix; the gene reconstruction loss is determined based on the genotype matrix and the reconstructed expression matrix;

[0080] Step 240: Calculate predicted phenotypic data of the genotype matrix by the prediction module; the phenotypic prediction loss is determined based on the phenotypic trait data and the predicted phenotypic data;

[0081] Step 250: Backpropagate the trait prediction model based on a loss function to update the network parameters of the trait prediction model; the loss function is determined based on the gene reconstruction loss and the phenotype prediction loss.

[0082] Specifically, the main contents of establishing a trait prediction model based on the target dataset include:

[0083] Preset the parameters of the multi-task variational autoencoder; initialize the parameters of each component of the multi-task variational autoencoder. In this embodiment, the Xavier method (by adjusting the variance of the weight initialization, the variance of the activation values ​​of each layer is kept consistent, thereby promoting the stability of forward / backward propagation) can be used for parameter initialization.

[0084] The genotype matrix As the model input, it is input to the task variational autoencoder, and the encoding module calculates the compressed expression of the genotype data. ;Generate module uses generator to generate randomly , and calculate the generated expression , is calculated as shown in Formula 1.

[0085] ; (1)

[0086] Calculate the reconstructed expression of genotype data through the decoding module ; Calculate predicted phenotypic data through the prediction module ; The calculation of the loss function is shown in formula 2, where, is the preset multi-task loss balance parameter; is the loss of the genotype data reconstruction task, which is calculated as shown in Formula 3; is the loss of phenotypic data prediction, which is calculated as shown in Formula 4.

[0087] ; (2)

[0088] ; (3)

[0089] ; (4)

[0090] in, for and The error can be calculated using the mean square error (MSE); is the KL divergence (Kullback-Leibler divergence) between the original genomic data and the generated expression; For the The prediction loss of a phenotypic trait can be calculated using the root mean square error (MSE) when the trait is predicted as a regression problem, or the cross entropy (CE) when the trait is predicted as a classification problem. express Distribution and KL divergence of the distribution; Indicates the mean , the standard deviation is Normal distribution; Represents the first phenotypic traits; Represents the first phenotypic traits; For the The model prediction data of phenotypic traits corresponds to a slice of For the Real data on phenotypic traits.

[0091] Backpropagation is performed on the multi-task variational autoencoder using the preset learning rate (LR) and loss function to update the network parameters. Repeat the above steps for multiple iterations according to the preset number of training steps and convergence requirements to obtain the trait prediction model.

[0092] This embodiment establishes a trait prediction model based on the target dataset by using a multi-task variational autoencoder.

[0093] In one embodiment, the breeding trait analysis method provided by the embodiment of the present invention may further include:

[0094] Step 111: determining gene locus data of each sample based on the genotype data of the breeding samples; the gene locus data includes identical homozygous sites, different homozygous sites, and heterozygous sites;

[0095] Step 112: Convert the gene site data into numerical data to obtain a genotype matrix, wherein the same homozygous sites, the different homozygous sites, and the heterozygous sites are converted into different numerical values.

[0096] Specifically, the preprocessing of genotype data includes the following steps: performing quality control on the genotype data to obtain quality-controlled gene sequencing data; selecting a reference genome and aligning the quality-controlled sequencing data; and detecting genetic variant sites from the comparison results. The genotype data for variant detection are shown in Table 3. Each row represents the single nucleotide polymorphism (SNP) information of a gene locus; the column represents the sample number. Genotype data can be in four forms: A, T, C, and G, and missing data are marked with NN.

[0097] Table 3

[0098]

[0099] Table 4

[0100]

[0101] The base data in Table 3 are converted to numerical data to form encoded genotype data. Specifically, the gene variant sites are compared with the reference genome, and homozygous sites that are identical to the reference genome are converted to 0, homozygous sites that are different from the reference genome are converted to 2, heterozygous sites are converted to 1, and missing values ​​are converted to -1. This yields the encoded genotype data shown in Table 4. The genotype matrix in this embodiment can be derived based on the encoded genotype data.

[0102] This embodiment converts the genotype data into a data format that is convenient for model training by performing pre-processing such as conversion processing on the genotype data.

[0103] In one embodiment, the breeding trait analysis method provided by the embodiment of the present invention may further include:

[0104] Step 310: determining a target variant site, a total set of variant sites, and a subset of variant sites based on the gene variant sites;

[0105] Step 320: Determine a first prediction result corresponding to the subset of variant sites using the trait prediction model;

[0106] Step 330: Determine a second prediction result corresponding to a target union using the trait prediction model; the target union is determined based on the target variant site and the variant site subset;

[0107] Step 340: Determine the effect matrix of the target variant site based on the total set of variant sites, the subset of variant sites, the first prediction result, and the second prediction result.

[0108] Specifically, the breeding trait analysis method provided in this embodiment uses the Shapley method to calculate the effect value of each gene variation node on each trait in the above trait prediction model.

[0109] Targeting gene mutation sites , i.e., the target variant site in this embodiment, is calculated by the trait prediction model using formula 5 Effect size on each trait .in, is the set of all gene variant sites in the genotype data, i.e., the total set of variant sites in this embodiment; yes Any subset of , i.e., the subset of variant sites in this embodiment; yes The number of gene variant sites in yes Eliminate variant sites The collection after is a collection Gene mutation sites The union of , that is, the target union in this embodiment; is the prediction function of the trait prediction model; That is, the second prediction result corresponding to the target union in this embodiment; is the genotype matrix according to Make slices; is the currently input prediction result, that is, the first prediction result corresponding to the variant site subset in this embodiment.

[0110] ; (5)

[0111] Obtained It is an N×K dimensional matrix, where N is the number of phenotypic traits and K is the number of samples in the target dataset. is the gene variation site of the Kth sample Character The effect value of .

[0112] Traverse the effect values ​​of all gene mutation sites, transpose these effect values ​​and integrate them into the effect matrix (shap matrix) of the gene mutation site, that is, the utility matrix in this embodiment, where: , as shown in the table.

[0113] Table 5

[0114]

[0115] This embodiment analyzes the effect of each gene variation site on each phenotypic trait through a fair contribution allocation method based on cooperative game theory, making full use of genotypic data and phenotypic data to improve the accuracy of breeding trait analysis.

[0116] In one embodiment, the breeding trait analysis method provided by the embodiment of the present invention may further include:

[0117] Step 410: Slice and extract the effect matrix according to the target trait to obtain a target matrix;

[0118] Step 420: Perform sample aggregation processing on the target matrix to obtain a first matrix;

[0119] Step 430: Perform phenotype aggregation processing on the first matrix to obtain a second matrix;

[0120] Step 440: Screen the gene variation sites based on the second matrix to obtain key variation sites corresponding to the target trait.

[0121] One or more traits are selected from the phenotypic traits as the target traits in this embodiment, and are expressed as a target trait set T in the form of a set; the effect matrix is ​​sliced ​​and extracted according to the target traits to obtain the target effect matrix , which is the target matrix in this embodiment; based on the sample, the target effect matrix is ​​aggregated, and the aggregated effect matrix is , which is the first matrix in this embodiment, the aggregation method is shown in Formula 6, where, is the absolute value function.

[0122] ; (6)

[0123] ; (7)

[0124] According to phenotypic traits After aggregation, the effect matrix is , which is the second matrix in this embodiment, the aggregation method is shown in Formula 7, where, is the maximum aggregation function. is: [0.000140, 0.000042, 0.000046, 0.000105, 0.000158].

[0125] right Arrange in descending order and select the first V gene variant sites with the largest effect values ​​as key variant sites. In the example of this embodiment, when V is 10, select The top 10 gene variant sites with the largest effect sizes are as follows: [X2_837053_C, X2_892379_T, X9_104863109_A, X8_3684564_A, X2_2067198_T, X2_1333237_C, X2_1427131_A, X9_108613741_A, X1_299312468_G, X5_209091560_T].

[0126] When there is only one target trait, the key variant sites screened out can be used as an auxiliary tool for genome-wide association analysis of the trait; when there are multiple target traits, the key variant sites screened out do not need to define individual trait weights, and are the analysis of the synergistic effects of multiple phenotypic traits by deep learning models.

[0127] This embodiment aggregates gene variation sites through the selected target traits to screen key variation sites with larger effect values, thereby providing support for different breeding goals.

[0128] The breeding trait analysis device provided by the present application is described below. The breeding trait analysis device described below can be referred to in correspondence with the breeding trait analysis method described above.

[0129] Please refer to Figure 3 The present application also provides a breeding trait analysis device, comprising:

[0130] A target data set construction module 301 is configured to construct a target data set based on genotype data and phenotype trait data of breeding samples.

[0131] A model training module 302 is configured to train a trait prediction model based on the target data set.

[0132] An effect matrix construction module 303 is configured to analyze the effect of a genetic variation site on a phenotype trait based on the trait prediction model to obtain an effect matrix; the genetic variation site is determined based on the genotype data.

[0133] A trait analysis module 304 is configured to aggregate the effect matrix to obtain a key variation site corresponding to a target trait; the target trait is selected from the phenotype trait data according to requirements.

[0134] Optionally, the target data set construction module comprises:

[0135] A data conversion unit is configured to convert the genotype data of the breeding samples to obtain a genotype matrix.

[0136] A data encoding unit is configured to encode the phenotype trait data of the breeding samples to obtain a trait encoding matrix.

[0137] A target data set construction unit is configured to integrate the genotype matrix and the trait encoding matrix based on the number of the breeding samples to obtain a target data set.

[0138] Optionally, the trait prediction model comprises an encoding module, a sampling module, a decoding module, and a prediction module; the model training module comprises:

[0139] A compressed expression matrix calculation unit is configured to take the genotype matrix as a model input, and calculate a compressed expression matrix of the genotype data based on the encoding module.

[0140] A generated expression matrix calculation unit is configured to calculate a generated expression matrix based on the sampling module and the compressed expression matrix.

[0141] A reconstructed expression matrix calculation unit is configured to calculate a reconstructed expression matrix based on the decoding module and the generated expression matrix; a gene reconstruction loss is determined based on the genotype matrix and the reconstructed expression matrix.

[0142] A predicted phenotypic data calculation unit, configured to calculate the predicted phenotypic data of the genotype matrix using the prediction module; the phenotypic prediction loss is determined based on the phenotypic trait data and the predicted phenotypic data;

[0143] A parameter updating unit is used to perform backpropagation on the trait prediction model based on a loss function to update the network parameters of the trait prediction model; the loss function is determined based on the gene reconstruction loss and the phenotype prediction loss.

[0144] Optionally, the data conversion unit includes:

[0145] A gene locus data determination unit, configured to determine the gene locus data of each sample based on the genotype data of the breeding sample; the gene locus data includes identical homozygous sites, different homozygous sites, and heterozygous sites;

[0146] The genotype matrix determining unit is used to convert the gene site data into numerical data to obtain a genotype matrix, wherein the same homozygous sites, the different homozygous sites and the heterozygous sites are converted into different numerical values.

[0147] Optionally, the effect matrix building module includes:

[0148] A variant site determination unit, configured to determine a target variant site, a total set of variant sites, and a subset of variant sites based on the gene variant sites;

[0149] A first prediction result determining unit, configured to determine a first prediction result corresponding to the subset of variant sites using the trait prediction model;

[0150] A second prediction result determining unit, configured to determine a second prediction result corresponding to a target union using the trait prediction model; the target union is determined based on the target variant site and the variant site subset;

[0151] An effect matrix determination unit is used to determine the effect matrix of the target variant site based on the total set of variant sites, the subset of variant sites, the first prediction result and the second prediction result.

[0152] Optionally, the trait analysis module includes:

[0153] a target matrix obtaining unit, configured to slice and extract the effect matrix according to the target trait to obtain a target matrix;

[0154] A first matrix obtaining unit, configured to perform sample aggregation processing on the target matrix to obtain a first matrix;

[0155] a second matrix obtaining unit, configured to perform phenotype aggregation processing on the first matrix to obtain a second matrix;

[0156] The key variation site obtaining unit is used to screen the gene variation sites based on the second matrix to obtain the key variation sites corresponding to the target trait.

[0157] Figure 4 An example of a physical structure diagram of an electronic device is shown below. Figure 4 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communications bus 540. The processor 510 may call logic instructions in the memory 530 to execute a breeding trait analysis method, which includes: constructing a target dataset based on genotype data and phenotypic trait data of breeding samples; obtaining a trait prediction model through training the target dataset; analyzing the effects of gene variant sites on phenotypic traits using the trait prediction model to obtain an effect matrix; the gene variant sites are determined based on the genotype data; and the effect matrix is ​​aggregated to obtain key variant sites corresponding to a target trait; the target trait is selected from the phenotypic trait data as required.

[0158] Furthermore, the logic instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0159] In another aspect, the present application also provides a computer program product comprising a computer program, which can be stored on a non-transitory computer-readable storage medium, and the computer program, when executed by a processor, enables a computer to perform the breeding trait analysis method provided by any of the above methods, which comprises: constructing a target data set based on genotype data and phenotype trait data of breeding samples; training a trait prediction model through the target data set; analyzing the effect of a genetic variation site on a phenotype trait through the trait prediction model to obtain an effect matrix; the genetic variation site is determined based on the genotype data; and performing aggregation processing on the effect matrix to obtain a key variation site corresponding to a target trait; the target trait is selected from the phenotype trait data according to requirements.

[0160] In another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program, when executed by a processor, enables a computer to perform the breeding trait analysis method provided by any of the above methods, which comprises: constructing a target data set based on genotype data and phenotype trait data of breeding samples; training a trait prediction model through the target data set; analyzing the effect of a genetic variation site on a phenotype trait through the trait prediction model to obtain an effect matrix; the genetic variation site is determined based on the genotype data; and performing aggregation processing on the effect matrix to obtain a key variation site corresponding to a target trait; the target trait is selected from the phenotype trait data according to requirements.

[0161] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0162] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus necessary general hardware platforms, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in terms of the contribution to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A breeding trait analysis method, characterized in that: include: Construct the target dataset based on the genotype data and phenotypic trait data of the breeding samples; Obtaining a trait prediction model through training of the target data set; Analyzing the effect of gene variation sites on phenotypic traits using the trait prediction model to obtain an effect matrix; the gene variation sites are determined based on the genotype data; Aggregating the effect matrix to obtain key variation sites corresponding to the target trait; The target trait is selected from the phenotypic trait data according to requirements; The target data set is constructed based on the genotype data and phenotypic trait data of the breeding samples, including: The genotype data of the breeding samples are transformed to obtain the genotype matrix; Encode the phenotypic trait data of the breeding samples to obtain the trait coding matrix; Based on the numbering of the breeding samples, the genotype matrix and the trait coding matrix are integrated to obtain a target data set; The trait prediction model is established on the basis of the target dataset through a multi-task variational autoencoder; the multi-task variational autoencoder includes an encoding module, a sampling module, a decoding module and a prediction module; the trait prediction model obtained by training the target dataset includes: The genotype matrix is ​​used as a model input, and the compressed expression matrix of the genotype data is calculated by the encoding module; Calculating and generating an expression matrix through the sampling module and the compressed expression matrix; The reconstructed expression matrix is ​​calculated by the decoding module and the generated expression matrix; the gene reconstruction loss is determined based on the genotype matrix and the reconstructed expression matrix; Calculating predicted phenotypic data of the genotype matrix by the prediction module; determining phenotypic prediction loss based on the phenotypic trait data and the predicted phenotypic data; Backpropagation is performed on the trait prediction model based on a loss function to update network parameters of the trait prediction model; the loss function is determined based on the gene reconstruction loss and the phenotype prediction loss; The effect matrix obtained by analyzing the effect of gene variant sites on phenotypic traits using the trait prediction model includes: Determine target variant sites, total variant site set, and variant site subset based on gene variant sites; Determining a first prediction result corresponding to the subset of variant sites using the trait prediction model; Determining a second prediction result corresponding to a target union using the trait prediction model; the target union is determined based on the target variant site and the variant site subset; Determining an effect matrix of the target variant site based on the total set of variant sites, the subset of variant sites, the first prediction result, and the second prediction result; By formula The effect matrix is ​​obtained; wherein, Gene mutation site Effect size on each trait; is the subset of variant sites; for Eliminate variant sites The collection after for the said target and set; is the prediction function of the trait prediction model; is the second prediction result; For the genotype matrix according to Made slices; is the first prediction result; is an N×K dimensional matrix, Transpose integration is performed to obtain the effect matrix; N is the number of phenotypic traits; and K is the number of samples in the target data set.

2. The breeding trait analysis method according to claim 1, characterized in that The genotype data of the breeding samples are converted to obtain a genotype matrix including: Determining gene locus data of each sample based on the genotype data of the breeding samples; the gene locus data includes identical homozygous sites, different homozygous sites, and heterozygous sites; The gene site data is converted into numerical data to obtain a genotype matrix, wherein the same homozygous sites, the different homozygous sites and the heterozygous sites are converted into different numerical values.

3. The breeding trait analysis method according to claim 1, wherein The key variation sites corresponding to the target trait obtained by aggregating the effect matrix include: Slice and extract the effect matrix according to the target trait to obtain a target matrix; Performing sample aggregation processing on the target matrix to obtain a first matrix; Performing phenotypic aggregation processing on the first matrix to obtain a second matrix; The gene variation sites are screened based on the second matrix to obtain key variation sites corresponding to the target trait.

4. A breeding trait analysis device, characterized in that: include: A target dataset construction module is used to construct a target dataset based on the genotype data and phenotypic trait data of breeding samples; A model training module, used to obtain a trait prediction model through training of the target data set; An effect matrix construction module, configured to analyze the effects of gene variant sites on phenotypic traits using the trait prediction model to obtain an effect matrix; the gene variant sites are determined based on the genotype data; A trait analysis module is used to aggregate the effect matrix to obtain key variation sites corresponding to the target trait; The target trait is determined based on the phenotypic trait data; The target data set construction module includes: A data conversion unit is used to convert the genotype data of the breeding sample to obtain a genotype matrix; A data encoding unit is used to encode the phenotypic trait data of the breeding sample to obtain a trait encoding matrix; a target data set construction unit, configured to integrate the genotype matrix and the trait coding matrix based on the number of the breeding samples to obtain a target data set; The trait prediction model is established on the basis of the target data set through a multi-task variational autoencoder; the multi-task variational autoencoder includes an encoding module, a sampling module, a decoding module and a prediction module; the model training module includes: a compressed expression matrix calculation unit, configured to use the genotype matrix as a model input and calculate the compressed expression matrix of the genotype data through the encoding module; An expression matrix calculation unit is used to calculate and generate an expression matrix through the sampling module and the compressed expression matrix; A reconstructed expression matrix calculation unit is used to calculate a reconstructed expression matrix through the decoding module and the generated expression matrix; the gene reconstruction loss is determined based on the genotype matrix and the reconstructed expression matrix; A predicted phenotypic data calculation unit, configured to calculate the predicted phenotypic data of the genotype matrix using the prediction module; the phenotypic prediction loss is determined based on the phenotypic trait data and the predicted phenotypic data; a parameter updating unit, configured to perform backpropagation on the trait prediction model based on a loss function, and update network parameters of the trait prediction model; the loss function is determined based on the gene reconstruction loss and the phenotype prediction loss; The effect matrix building block includes: A variant site determination unit, configured to determine a target variant site, a total set of variant sites, and a subset of variant sites based on the gene variant sites; A first prediction result determining unit, configured to determine a first prediction result corresponding to the subset of variant sites using the trait prediction model; A second prediction result determining unit, configured to determine a second prediction result corresponding to a target union using the trait prediction model; the target union is determined based on the target variant site and the variant site subset; an effect matrix determining unit, configured to determine an effect matrix of the target variant site based on the total set of variant sites, the subset of variant sites, the first prediction result, and the second prediction result; By formula The effect matrix is ​​obtained; wherein, Gene mutation site Effect size on each trait; is the subset of variant sites; for Eliminate variant sites The collection after for the said target union; is the prediction function of the trait prediction model; is the second prediction result; For the genotype matrix according to Made slices; is the first prediction result; is an N×K dimensional matrix, Transpose integration is performed to obtain the effect matrix; N is the number of phenotypic traits; and K is the number of samples in the target data set.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the breeding trait analysis method according to any one of claims 1 to 3 is implemented.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the breeding trait analysis method according to any one of claims 1 to 3 is implemented.

7. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the breeding trait analysis method according to any one of claims 1 to 3 is implemented.

Images