A method, apparatus, medium, and program product for predicting hybrid phenotypes

Abstract

This application provides a method, device, medium, and program product for predicting hybrid phenotypes, relating to the field of breeding technology. The method includes: acquiring genotype data of the offspring of the test subject at multiple loci; inputting the genotype data of the offspring of the test subject at multiple loci into a target phenotype prediction model to obtain predicted phenotype data of the test subject; wherein the target phenotype prediction model is the model with the highest prediction accuracy among non-parametric prediction models, parametric prediction models, and semi-parametric prediction models. During the training process of the prediction model, the training data is specifically processed. The target phenotype prediction model in this application can more accurately predict the phenotype data of the test subject.

Description

Technical Field

[0001] This application relates to the field of breeding technology, specifically to a method, equipment, medium, and program product for predicting hybrid phenotypes. Background Technology

[0002] Selecting superior hybrids with high yield potential is one of the important tasks of breeding. Superior hybrids with high yield potential can increase yield by more than 20% by utilizing heterosis.

[0003] Conventional hybridization breeding typically involves using superior inbred lines as parents to generate a large number of hybrids. These hybrids are then subjected to field trials in various environments to determine their yield under different conditions. Finally, high-yielding hybrids are selected based on their yield performance. This method, however, is limited by human, material, and land resources, resulting in low efficiency in selecting superior hybrids. With the development of breeding technology, Genomic Selection (GS) has emerged. GS uses modeling to simulate and predict the phenotypic data of all possible hybrids, selecting high-yielding hybrids based on this data. This method eliminates the need for field trials in various environments, saving resources and significantly increasing the efficiency of selecting superior hybrids.

[0004] However, current genome-wide selection technology has the drawback of low accuracy in predicting the phenotypic data of hybrids. Summary of the Invention

[0005] The purpose of this application is to address the above-mentioned problems by providing a method, device, medium, and program product for predicting hybrid phenotypes, so as to improve the accuracy of phenotypic data predicted by whole-genome selection technology.

[0006] In a first aspect, this application provides a method for predicting the phenotype of hybrids, comprising:

[0007] Obtain genotype data of the offspring of the subject at multiple loci;

[0008] The genotype data of the offspring of the test subject at multiple loci are input into the target phenotype prediction model to obtain the predicted phenotype data of the test subject.

[0009] The target phenotype prediction model is selected from non-parametric prediction models, parametric prediction models, and semi-parametric prediction models based on the highest prediction accuracy. The non-parametric prediction model is trained using a first-generation genotype data matrix and corresponding first-phenotype data. The first-generation genotype data matrix represents the genotype data of multiple generations at multiple loci. The parametric prediction model is trained using a second-generation genotype data matrix and second-phenotype data. The second-generation genotype data matrix is ​​obtained based on the first-generation genotype data matrix and a dominance matrix, which is a matrix composed of the interaction relationships between different alleles at the same locus. The semi-parametric prediction model is trained using a third-generation genotype data matrix and third-phenotype data. The third-generation genotype data matrix is ​​obtained based on the first-generation genotype data matrix and Euclidean distance, which represents the similarity of gene data at multiple loci of any two generations.

[0010] In some possible implementations, the first offspring genotype data matrix is ​​determined in the following way:

[0011] Obtain the fourth-generation offspring genotype data matrix;

[0012] The genotype data matrix of the fourth offspring is filtered to obtain the genotype data matrix of the first offspring.

[0013] In some possible implementations, the step of filtering the fourth-generation genotype data matrix to obtain the first-generation genotype data matrix includes:

[0014] Obtain the correlation between genotype data and fourth phenotype data at each locus in the fourth offspring genotype data matrix;

[0015] The sites containing genotype data with correlation below a preset threshold are deleted to obtain the first generation genotype data matrix.

[0016] In some possible implementations, the target phenotype prediction model is determined in the following ways:

[0017] Obtain the first accuracy of the nonparametric prediction model, the second accuracy of the parametric prediction model, and the third accuracy of the semiparametric prediction model;

[0018] The maximum value among the first accuracy, second accuracy, and third accuracy is determined as the target accuracy;

[0019] The model corresponding to the target accuracy is used as the target phenotype prediction model.

[0020] In some possible implementations, the method further includes:

[0021] Field planting experiments were conducted on the offspring of the tested subjects to obtain experimental phenotypic data;

[0022] The experimental phenotypic data and the predicted phenotypic data are compared to obtain the comparison results;

[0023] If the comparison results indicate that the difference between the experimental phenotype data and the predicted phenotype data is greater than a preset difference threshold, a prompt message is generated; the prompt message is used to suggest reselecting the target phenotype prediction model or updating the target phenotype prediction model.

[0024] In some possible implementations, obtaining genotypic data of the offspring of the test subject at multiple loci includes:

[0025] Genotype data of the test subject at multiple loci are obtained, wherein the test subject is the father and mother of the offspring;

[0026] Based on the genotype data of the test subject at multiple loci, the genotype data of the offspring of the test subject at multiple loci are inferred.

[0027] Secondly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above.

[0028] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0029] Fourthly, this application provides a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the process or function as described above is generated.

[0030] Compared with existing technologies, the beneficial effects of this application are as follows: In this application, the training data is first processed specifically during the training of the prediction model. Specifically, based on the first generation genotype data matrix and corresponding phenotypic data, the influence of dominant results in heterosis is considered to generate a second generation genotype data matrix for training the parametric prediction model. The mutual influence between samples is considered to generate a third generation genotype data matrix for training the semi-parametric prediction model. Furthermore, a feature filtering method is used to remove loci with low correlation to the phenotypic data. Then, the prediction model with the highest accuracy is selected from the multiple trained prediction models as the target phenotypic prediction model. Finally, the genotype data of the offspring of the test object at multiple loci are input into the target phenotypic prediction model to obtain the predicted phenotypic data of the offspring of the test object. Since the target phenotypic prediction model has the highest accuracy, and the above three models are all trained using processed data, the training time of the prediction model is shortened. Therefore, the target phenotypic prediction model in this application can more accurately predict the phenotypic data of the test object, while reducing the prediction cost.

[0031] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in this embodiment, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A flowchart illustrating a method for predicting hybrid phenotypes provided in this application embodiment;

[0034] Figure 2 A box plot for predicting the accuracy of hybrid phenotypic data provided in this application embodiment;

[0035] Figure 3 A schematic diagram illustrating the training process of a target phenotype prediction model provided in an embodiment of this application;

[0036] Figure 4 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0037] To enable those skilled in the art to better understand the technical solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. The descriptions in this section are merely illustrative and explanatory, and should not be construed as limiting the scope of protection of this application. Specifically, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort should fall within the scope of protection of this application.

[0038] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," and "third," etc., in this specification and the description of the figures are used to distinguish different objects, not to define a specific order. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes 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 these processes, methods, products, or devices.

[0039] As mentioned in the background section, this application provides a method for predicting the phenotype of hybrids to address the problems in the prior art. The method includes:

[0040] Genotypic data of the offspring of the test subject at multiple loci are obtained; a first offspring genotypic data matrix and corresponding first phenotype data representing the genotypic data of multiple offspring at multiple loci are obtained; a second offspring genotypic data matrix is ​​constructed based on the first offspring genotypic data matrix and the dominance matrix, and a third offspring genotypic data matrix is ​​constructed based on the first offspring genotypic data matrix and the Euclidean distance; then, a parametric prediction model is trained using the second offspring genotypic data matrix and the second phenotype data, a semi-parametric prediction model is trained using the third offspring genotypic data matrix and the third phenotype data, and a non-parametric prediction model is trained using the first offspring genotypic data matrix and the first phenotype data; the model with the highest accuracy among the non-parametric, parametric, and semi-parametric prediction models is selected as the target phenotype prediction model; the genotypic data of the offspring of the test subject at multiple loci are input into the target phenotype prediction model to obtain the predicted phenotype data of the test subject.

[0041] In this application, the training data is first processed specifically during the training of the prediction model. Specifically, based on the first-generation genotype data matrix and corresponding phenotypic data, the influence of dominant results in heterosis is considered to generate a second-generation genotype data matrix for training the parametric prediction model. Considering the mutual influence between samples, a third-generation genotype data matrix is ​​generated for training the semi-parametric prediction model. Furthermore, a feature filtering method is used to remove loci with low correlation to the phenotypic data. Then, the prediction model with the highest accuracy is selected from the multiple trained prediction models as the target phenotypic prediction model. Finally, the genotype data of the offspring of the test object at multiple loci are input into the target phenotypic prediction model to obtain the predicted phenotypic data of the offspring of the test object. Since the target phenotypic prediction model has the highest accuracy, and the above three models are all trained using processed data, the training time of the prediction model is shortened. Therefore, the target phenotypic prediction model in this application can more accurately predict the phenotypic data of the test object, while reducing the prediction cost.

[0042] To make the technical solution of this application clearer and easier to understand, the following describes a method for predicting the phenotype of a hybrid species provided in the embodiments of this application.

[0043] like Figure 1 As shown in the figure, this is a flowchart of a method for predicting the phenotype of a hybrid provided in this embodiment. The method includes the following steps:

[0044] S101. Obtain genotype data of the offspring of the subject under test at multiple loci.

[0045] In some embodiments, a site refers to a single nucleotide polymorphism site.

[0046] The test subjects refer to the male and female parents of the offspring in hybridization breeding. Offspring are obtained by crossing the male and female parents. There can be multiple test subjects in this application, and there can also be multiple corresponding offspring. Locus refers to the specific position of a gene on a chromosome, and genotype refers to the type of gene combination that controls a trait. The genotype data of the offspring is usually inferred from the genotype data of the parents. The specific method for obtaining the data is as follows:

[0047] In some embodiments, obtaining genotype data of the offspring of the subject at multiple loci includes:

[0048] Genotypic data of the test subjects at multiple loci are obtained, with the test subjects being the father and mother of the offspring; genotypic data of the offspring at multiple loci are inferred based on the genotypic data of the test subjects at multiple loci.

[0049] For example, if the genotype of the father (P1) at a certain locus is P1=AA, the genotype of the mother (P2) at the corresponding locus is P2=aa, and the genotype of the first generation (F1) is Aa.

[0050] To avoid ambiguity, it is specifically stated that the genotype data used in predicting phenotypic data in this application is only for the first generation (F1). Accordingly, the terms "first," "second," and "third" mentioned later are used to distinguish different objects, rather than to limit them to F1, F2, F3, etc.

[0051] To facilitate prediction and calculation, genotypes are replaced with numbers. In this application, genotype data at multiple loci are represented as vectors or matrices. The padding data for heterozygous genotypes is 0, for homozygous dominant genotypes it is -1, and for homozygous recessive genotypes it is 1. Other substitution relationships can be used in practice, and this application does not limit this. For example, if the offspring (F1) have genotypes AA, Aa, and aa at three loci, the corresponding genotype data would be: Accordingly, genotypic data for multiple offspring at multiple loci are... The matrix is ​​given by M, where M represents the number of offspring and N represents the number of loci.

[0052] S102. Input the genotype data of the offspring of the test subject at multiple loci into the target phenotype prediction model to obtain the predicted phenotype data of the test subject.

[0053] Phenotype refers to the structural and functional characteristics of an organism that can be observed. For example, for wheat, the height of the plant, the fullness of the ears, and its resistance to lodging are all phenotypic data of wheat.

[0054] First, obtain the target phenotype prediction model, and then input the genotype data of the offspring of the test subject at multiple loci into the target phenotype prediction model to obtain the predicted phenotype data of the test subject.

[0055] The target phenotype prediction model will be introduced in detail below. The target phenotype prediction model is the model with the highest accuracy in predicting phenotype data among nonparametric prediction models, parametric prediction models and semiparametric prediction models.

[0056] Parametric, semi-parametric, and non-parametric prediction models are the core of genome-wide selection (WWC) technology, significantly impacting its accuracy and efficiency. Parametric prediction models include Genomic Best Linear Unbiased Prediction (GBLUP), Ridge Regression Best Linear Unbiased Prediction (rrBLUP), Bayesian A, Bayesian B, Bayesian C, Bayesian Lasso, and Bayesian Ridge Regression (BRR). Semi-parametric prediction models include Reproducing Kernel Hilbert (RKHS). Non-parametric prediction models encompass various machine learning models. The process of using prediction models involves training and application. The corresponding data is divided into training and prediction sets. The training set requires both phenotypic and genotypic data for training the prediction model, while the prediction set only requires genotypic data to predict phenotypes based on the genotypic data and the prediction model.

[0057] The nonparametric prediction model is trained using the first offspring genotype data matrix and the corresponding first phenotype data. The first offspring genotype data matrix represents the genotype data of multiple offspring at multiple loci. When the first offspring genotype data matrix represents the genotype data of M offspring at N loci.

[0058] The corresponding first phenotypic data can be one or more. Taking one phenotypic as an example, the technical solution of this application is introduced, such as plant height.

[0059] The parameter prediction model is trained using the second-generation genotype data matrix and the second-generation phenotype data. The second-generation genotype data matrix is ​​based on the first-generation genotype data matrix and the dominance matrix. The dominance matrix is ​​a matrix composed of the interaction relationships between different alleles at the same locus.

[0060] The data used in training the parametric prediction model is the same as that used in the non-parametric prediction model; that is, the second phenotypic data is the same as the first phenotypic data, which will not be elaborated here. The difference lies in the offspring genotype data. The second offspring genotype data matrix is ​​described below:

[0061] Set the first offspring genotype data matrix as follows: The matrix, where M represents the number of offspring and N represents the number of loci, is used for parametric prediction models such as rrBLUP and Bayesian methods (BayesA, BayesB, BayesC, Bayes Lasso, and BRR). The dominant matrix is ​​represented as follows: The dominant eigenvalue matrix, where the filling element is D. mn D mn The value represents whether the genotype at the nth locus of the mth offspring is heterozygous. For example, heterozygous is marked as 1, and the rest are marked as 0. The dominant trait matrix is ​​concatenated with the original first offspring genotype matrix to form the second offspring genotype data matrix, which is used as input data to the above prediction model. The purpose is to distinguish dominant traits, fully consider heterosis, and make the prediction results more accurate.

[0062] For example, when M=4 and N=3, the genotype data matrix of the first offspring is as follows:

[0063]

[0064] The corresponding second-generation genotype data matrix is ​​as follows:

[0065]

[0066] It should be noted that, because GBLUP uses individuals as random covariates in the parametric prediction model, unlike other prediction models that use loci as random covariates, a different form of input matrix is ​​used for GBLUP: additive effects matrix and dominant effects matrix. These matrices are calculated based on the first-generation progeny genotype data matrix. The additive effects matrix is... The matrix is ​​filled with elements A. mm A mm The dominant effect matrix represents the cumulative value of multiple minor gene effects. The matrix is ​​filled with elements D. mm D mm The values ​​representing the interactions between different alleles at the same locus are input data using the additive effect matrix and the dominant effect matrix, and are also input into the GBLUP model. The aim is to fully consider heterosis and make the prediction results more accurate.

[0067] The semi-parametric prediction model is trained using the third-generation genotype data matrix and the third-generation phenotype data. The third-generation genotype data matrix is ​​obtained based on the first-generation genotype data matrix and the Euclidean distance, which represents the similarity of gene data at multiple loci in any two generations.

[0068] The data used in the training process of the semi-parametric prediction model is the same as that used in the non-parametric prediction model. That is, the third phenotype data is the same as the first phenotype data, which will not be repeated here. The difference lies in the offspring genotype data. The third offspring genotype data matrix is ​​introduced below.

[0069] Based on the same first-generation genotype matrix, and considering the similarity between samples, the Euclidean distance is used to calculate the final third-generation genotype data matrix. The third-generation genotype data is as follows: The matrix is ​​filled with elements S. ij , of which S ij The value represents the magnitude of genotypic similarity between the i-th and j-th offspring at multiple loci.

[0070] In some embodiments, the target phenotype prediction model is determined in the following ways:

[0071] Obtain the first accuracy of the nonparametric prediction model, the second accuracy of the parametric prediction model, and the third accuracy of the semiparametric prediction model; determine the maximum value among the first, second, and third accuracies as the target accuracy; and use the model corresponding to the target accuracy as the target phenotype prediction model.

[0072] To make better use of the data in the training set and improve the accuracy of the prediction model, this application employs the K-fold cross-validation method, using part of the data in the training set for training and part for accuracy validation. Specifically, the process of obtaining the first accuracy of the nonparametric prediction model is described below:

[0073] Obtain a validation set, which contains genotype and phenotypic data of multiple offspring at multiple loci in the offspring genotype data matrix; use the validation set to validate the accuracy of the trained nonparametric prediction model and obtain the first accuracy.

[0074] The methods for obtaining the second and third accuracies differ from those for obtaining the first accuracies in that they use different validation sets. This application employs different training and validation sets for parametric prediction models, semi-parametric prediction models, and non-parametric prediction models, thereby improving the validation accuracy and making the accuracy more reliable.

[0075] After obtaining the first accuracy, second accuracy, and third accuracy, the maximum value among the first accuracy, second accuracy, and third accuracy is determined as the target accuracy, and the model corresponding to the target accuracy is used as the target phenotype prediction model.

[0076] In some embodiments, the first-generation genotype data matrix is ​​determined in the following manner:

[0077] Obtain the genotype data matrix of the fourth generation; filter the genotype data matrix of the fourth generation to obtain the genotype data matrix of the first generation.

[0078] The fourth-generation genotype data matrix contains genotype data from multiple generations at multiple loci. Compared with the first-generation genotype data, there are a large number of loci with little or no correlation to the phenotype.

[0079] In some embodiments, the fourth-generation genotype data matrix is ​​filtered to obtain the first-generation genotype data matrix, including:

[0080] Obtain the correlation between genotype data and fourth phenotype data at each locus in the fourth offspring genotype data matrix; delete the loci containing genotype data with correlation below a preset threshold to obtain the first offspring genotype data matrix.

[0081] Correlation represents the magnitude of the correlation between genotype and phenotypic data at a given locus. A greater correlation indicates a stronger correlation, and a smaller correlation indicates a weaker correlation. To improve phenotypic prediction accuracy and shorten training time, loci with low correlation to phenotypic data in the fourth-generation genotype data matrix were removed, resulting in the first-generation genotype data matrix.

[0082] Specifically, in order to remove loci with low correlation to phenotypic data as much as possible, different filtering feature methods are selected for different prediction models. The filtering feature methods provided in this application include F-test, mutual information and wrapping method. Other filtering feature methods can also be selected. This application does not impose any restrictions. The appropriate method can be selected as needed in actual operation.

[0083] The F-test is a filtering method used to capture the linear relationship between each feature and the label; mutual information is used to capture any linear and non-linear relationship between each feature and the label phenotypic data; the wrapping method (RFECV) is a combination of recursive feature elimination (PRE) and cross-validation (CV). The operating mechanism is to first use PRE to obtain the weight ranking of each feature, and then select a predetermined number of feature subsets for model training and cross-validation based on the weight ranking, and finally select the feature subset with the highest average score.

[0084] Since the parametric prediction model is a linear relationship model, this application uses the F-test method for site deletion, while the semi-parametric prediction model contains non-linear relationships, so this application uses the mutual information method, and the non-parametric prediction model uses the wrapping method for site deletion.

[0085] This application selects the F-test method for the parameter prediction model, and the specific process is as follows:

[0086] Obtain first feature data and first label data. The first feature data is the offspring genotype data matrix corresponding to the parameter prediction model, and the first label data is the phenotype corresponding to the offspring genotype data matrix. Input the genotype data and first label data of the locus to be tested in the first feature data into the F-test model in sequence to obtain the F value of the locus to be tested. The F value represents the correlation value between the genotype data and the label data of the locus to be tested. If the F value meets the preset F value range, the locus to be tested in the first feature data is retained; if the F value does not meet the preset F value range, the locus to be tested in the first feature data is deleted.

[0087] The mutual information method is selected for the semi-parametric prediction model, and the specific process is as follows:

[0088] Acquire second feature data and second label data. The second feature data is the offspring genotype data matrix corresponding to the semi-parametric prediction model, and the second label data is the phenotype corresponding to the offspring genotype data matrix. Input the genotype data and second label data of the locus to be tested in the second feature data into the mutual information formula to obtain the probability value of the locus to be tested. The probability value represents the correlation between the genotype data of the locus to be tested and the second label data. If the probability value meets the preset probability value range, the locus to be tested in the second feature data is retained; if the probability value does not meet the preset probability value range, the locus to be tested in the second feature data is deleted.

[0089] The specific process for selecting the wrapping method for nonparametric prediction models is as follows:

[0090] Obtain the third feature data and the third label data. The third feature data is the offspring genotype data matrix corresponding to the nonparametric prediction model, and the third label data is the phenotype corresponding to the offspring genotype data matrix. Obtain the weight ranking of the genotype data at each locus of the third feature data. The weight represents the correlation between the genotype data and the phenotype data at the locus. According to the weight ranking, select a predetermined number of loci and their genotype data as subsets in sequence, calculate the score of the subset, and the score represents the degree of correlation between the subset and the phenotype data. Compare the scores of each subset. Select the subset with the highest score as the first offspring genotype data matrix.

[0091] Figure 2 A box plot for predicting the accuracy of hybrid phenotypic data is provided in an embodiment of this application, such as... Figure 2 As shown in the figure, this graph represents the accuracy of the prediction model before and after filtering for phenotype P. The vertical axis represents the model name, and the horizontal axis represents the accuracy. The models include: parametric prediction models, semi-parametric prediction models, and non-parametric prediction models. The parametric prediction models include: Genomic Best Linear Unbiased Prediction (GBLUP), Ridge Regression Best Linear Unbiased Prediction (rrBLUP), Bayes A, Bayes B, Bayes C, Bayes Lasso, and Bayes Ridge Regression (BRR). The semi-parametric prediction models include: Reproducing Kernel Hilbert (RKHS). The non-parametric prediction models include: Support Vector Regression (SVR), Random Forest, and Lightweight Gradient Boosting. (Machine, LightGBM), no_filter_ indicates the prediction model before filtering, filter_ indicates the prediction model after filtering, where no_filter_GBLUP is the GBLUP model before filtering with an accuracy of 0.668, and filter_GBLUP is the GBLUP model after filtering with an accuracy of 0.661.

[0092] As shown in the figure, the accuracy of the parametric prediction model, semi-parametric prediction model, and non-parametric prediction model after filtering is equal to or higher than the accuracy before filtering. SVR has the highest accuracy and the greatest improvement in accuracy, with no_filter_SVR accuracy at 0.658 and filter_SVR accuracy at 0.769, representing an accuracy improvement of 10%-20%. For phenotype P, the target phenotype prediction model is SVR. This demonstrates that using the site deletion method to remove sites with low correlation to phenotype data can improve the prediction accuracy of the prediction model. Furthermore, taking phenotype P as an example, the training time of the RandomForest prediction model was reduced from 12 hours to two and a half hours, significantly reducing costs.

[0093] In some embodiments, the method further includes:

[0094] Field planting experiments are conducted on the offspring of the test subjects to obtain experimental phenotypic data. The experimental phenotypic data is compared with the predicted phenotypic data to obtain comparison results. For example, the experimental phenotypic data and the predicted phenotypic data can be scored for satisfaction, with higher scores for more satisfactory phenotypic data. If the comparison results indicate that the difference between the experimental phenotypic data and the predicted phenotypic data is greater than a preset difference threshold, a prompt message is generated. The prompt message is used to suggest reselecting the target phenotypic prediction model or updating the target phenotypic prediction model. The update can be to replace the training set used during training, optimize the prediction model itself, or use other methods to optimize the target phenotypic data.

[0095] Taking the phenotype of whether wheat ears are full as an example, the prediction model shows that the seed phenotype is full ears, with a fullness score of 10. After the seed is planted in the field and the wheat matures, the fullness score of the wheat ears is 7. The preset difference threshold is 2. If the actual difference is greater than the preset difference threshold, it means that other target phenotype prediction models need to be selected, or other training sets need to be used to train the target phenotype prediction model.

[0096] In some embodiments, the nonparametric prediction model employs the optimal combination of hyperparameters. This application uses Optuna to optimize the hyperparameters of the nonparametric prediction model. Optuna is a Python-based automatic hyperparameter optimization framework, the purpose of which is to improve the generalization ability of the nonparametric prediction model.

[0097] In some embodiments, the genotype data of the offspring of the test subject at multiple loci are preprocessed. If the target phenotype prediction model is a parametric prediction model, the preprocessing includes filtering the genotype data of the offspring of the test subject at multiple loci and combining the genotype data of the offspring of the test subject at multiple loci with the dominance matrix. If the target phenotype prediction model is a semi-parametric prediction model, the preprocessing package filters the test genotype data matrix and calculates and replaces the Euclidean distance of the genotype data of the offspring of the test subject at multiple loci. If the target phenotype prediction model is a non-parametric prediction model, the preprocessing includes filtering the test genotype data matrix. The purpose of preprocessing is to optimize the data input to the target phenotype prediction model, so as to improve the accuracy of the predicted phenotype data. The specific filtering, combination with the dominance matrix, and Euclidean distance calculation have been described in detail in the aforementioned training process, and will not be repeated here for the sake of brevity.

[0098] In this application, the training data is first processed specifically during the training of the prediction model. Specifically, based on the first-generation genotype data matrix and corresponding phenotypic data, the influence of dominant results in heterosis is considered to generate a second-generation genotype data matrix for training the parametric prediction model. Considering the mutual influence between samples, a third-generation genotype data matrix is ​​generated for training the semi-parametric prediction model. Furthermore, a feature filtering method is used to remove loci with low correlation to the phenotypic data. Then, the prediction model with the highest accuracy is selected from the multiple trained prediction models as the target phenotypic prediction model. Finally, the genotype data of the offspring of the test object at multiple loci are input into the target phenotypic prediction model to obtain the predicted phenotypic data of the offspring of the test object. Since the target phenotypic prediction model has the highest accuracy, and the above three models are all trained using processed data, the training time of the prediction model is shortened. Therefore, the target phenotypic prediction model in this application can more accurately predict the phenotypic data of the test object, while reducing the prediction cost.

[0099] In some embodiments, the training process of the target phenotype prediction model is as follows: Figure 3 As shown in the figure, this is a schematic diagram of the training process of a target phenotype prediction model provided in an embodiment of this application, including:

[0100] S301. Obtain the first offspring genotype data matrix and the corresponding first phenotype data. The first offspring genotype data matrix represents the genotype data of multiple offspring at multiple loci.

[0101] S302. Combine the first generation genotype data matrix with the dominance matrix to obtain the second generation genotype data matrix. The dominance matrix is ​​a matrix composed of the interaction relationships between different alleles at the same locus.

[0102] S303. Obtain the third generation genotype data matrix based on the first generation genotype data matrix and the Euclidean distance. The Euclidean distance represents the similarity of gene data at multiple loci in any two generations.

[0103] S304. The initial nonparametric prediction model is trained using the first offspring genotype data matrix and the first phenotype data to obtain the trained nonparametric prediction model.

[0104] S305. The initial parameter prediction model is trained using the second offspring genotype data matrix and the second phenotype data to obtain the trained parameter prediction model.

[0105] S306. The initial semi-parametric prediction model is trained using the third-generation progeny genotype data matrix and the third-generation phenotype data to obtain the trained semi-parametric prediction model.

[0106] The second phenotypic data and the third phenotypic data are consistent with the first phenotypic data.

[0107] The specific training process of the model training method provided in this application embodiment has been described in detail in the foregoing embodiments, and will not be repeated here for the sake of brevity.

[0108] The above text combined Figures 1 to 3 This application provides a detailed description of a method for predicting the phenotype of a hybrid. The electronic device and computer-readable storage medium provided in this application will be described below with reference to the accompanying drawings.

[0109] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above embodiments.

[0110] like Figure 4 As shown, this figure is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Figure 4 As shown, the computing device 400 includes a bus 401, a processor 402, a communication interface 403, and a memory 404. The processor 402, the memory 404, and the communication interface 403 communicate with each other via the bus 401.

[0111] Bus 401 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0112] Processor 402 can be any one or more of the following processors: central processing unit (CPU), graphics processing unit (GPU), microprocessor (MP), or digital signal processor (DSP).

[0113] Communication interface 403 is used for communication with external devices.

[0114] Memory 404 may include volatile memory, such as random access memory (RAM). Memory 404 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0115] The memory 404 stores executable code, which the processor 402 executes to perform the aforementioned method for predicting hybrid phenotypes.

[0116] Specifically, in achieving Figure 1 In the case of the illustrated embodiment, and Figure 1 The method for predicting hybrid phenotypes described in the embodiments is implemented in software, and involves executing... Figure 1 The required software or program code can be stored partially or entirely in memory 404. Processor 402 executes the program code corresponding to each unit stored in memory 404 to perform the aforementioned method for predicting hybrid phenotypes.

[0117] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium capable of being stored by a computing device, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute the aforementioned method for predicting hybrid phenotypes.

[0118] This application also provides a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application are generated.

[0119] The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another website, computer, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0120] When the computer program product is executed by a computer, the computer performs any of the aforementioned methods for predicting hybrid phenotypes. The computer program product can be a software installation package; when any of the aforementioned methods for predicting hybrid phenotypes needs to be used, the computer program product can be downloaded and executed on the computer.

[0121] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0122] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A method for predicting the phenotype of a hybrid, characterized in that, include: Obtain genotype data of the offspring of the subject at multiple loci; The genotype data of the offspring of the test subject at multiple loci are input into the target phenotype prediction model to obtain the predicted phenotype data of the test subject. Among them, the target phenotype prediction model is the model with the highest prediction accuracy selected from nonparametric prediction models, parametric prediction models and semiparametric prediction models; The nonparametric prediction model is trained using the first progeny genotype data matrix and the corresponding first phenotype data, wherein the first progeny genotype data matrix represents the genotype data of multiple progeny at multiple loci. The parameter prediction model is trained from the second progeny genotype data matrix and the second phenotype data. The second progeny genotype data matrix is ​​obtained based on the first progeny genotype data matrix and the dominance matrix. The dominance matrix is ​​a matrix composed of the interaction relationships between different alleles at the same locus. The semi-parametric prediction model is trained using the third-generation genotype data matrix and the third-generation phenotype data. The third-generation genotype data matrix is ​​obtained based on the first-generation genotype data matrix and the Euclidean distance. The Euclidean distance represents the similarity of gene data at multiple loci in any two generations. The first offspring genotype data matrix was determined in the following way: Obtain the fourth-generation offspring genotype data matrix; Obtain the correlation between genotype data and fourth phenotype data at each locus in the fourth-generation genotype data matrix; delete the loci containing genotype data with correlation below a preset threshold to obtain the first-generation genotype data matrix; correlation represents the magnitude of the correlation between genotype data and phenotype data at a locus, the greater the correlation with phenotype data, the stronger the correlation, and the smaller the correlation with phenotype data, the weaker the correlation; different filtering feature methods are selected for locus deletion for different prediction models; Among them, the F-test method is selected for site deletion for parametric prediction models, the mutual information method is selected for site deletion for semi-parametric prediction models, and the wrapping method is selected for site deletion for non-parametric prediction models. The method further includes: preprocessing the genotype data of the offspring of the test subject at multiple loci; if the target phenotype prediction model is a parametric prediction model, the preprocessing includes screening the genotype data of the offspring of the test subject at multiple loci and combining the genotype data of the offspring of the test subject at multiple loci with the dominance matrix; if the target phenotype prediction model is a semi-parametric prediction model, the preprocessing package includes screening the test genotype data matrix and calculating and replacing the Euclidean distance of the genotype data of the offspring of the test subject at multiple loci; if the target phenotype prediction model is a non-parametric prediction model, the preprocessing includes screening the test genotype data matrix. The second and third phenotypic data are consistent with the first phenotypic data.

2. The method according to claim 1, characterized in that, The screening of the fourth offspring genotype data matrix Selecting, we obtain the first generation genotype data matrix, including: Obtain the correlation between genotype data and fourth phenotype data at each locus in the fourth offspring genotype data matrix; The sites containing genotype data with correlation below a preset threshold are deleted to obtain the first generation genotype data matrix.

3. The method according to claim 1, characterized in that, The target phenotype prediction model is determined in the following way: Obtain the first accuracy of the nonparametric prediction model, the second accuracy of the parametric prediction model, and the third accuracy of the semiparametric prediction model; The maximum value among the first accuracy, second accuracy, and third accuracy is determined as the target accuracy; The model corresponding to the target accuracy is used as the target phenotype prediction model.

4. The method according to claim 1, characterized in that, The method further includes: Field planting experiments were conducted on the offspring of the tested subjects to obtain experimental phenotypic data; The experimental phenotypic data and the predicted phenotypic data are compared to obtain the comparison results; If the comparison results indicate that the difference between the experimental phenotype data and the predicted phenotype data is greater than a preset difference threshold, a prompt message is generated; the prompt message is used to suggest reselecting the target phenotype prediction model or updating the target phenotype prediction model.

5. The method according to claim 1, characterized in that, The acquisition of genotype data of the offspring of the test subject at multiple loci includes: Genotype data of the test subject at multiple loci are obtained, wherein the test subject is the father and mother of the offspring; Based on the genotype data of the test subject at multiple loci, the genotype data of the offspring of the test subject at multiple loci are inferred.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.

8. A computer program product, characterized in that, The computer program product includes instructions that, when executed by a computer, enable the computer to perform the method as described in any one of claims 1-5.

Images