A genome prediction method, prediction system and application

Abstract

The present invention belongs to the technical fields of genomics, bioinformatics, genome prediction and genome breeding, and discloses a genome prediction method, prediction system and application. The genome prediction method includes obtaining genotype information; constructing a genome prediction model; estimating the effect value of each genotype; sorting all genotypes from small to large according to the size of each genotype effect value, and gradually discarding the genotype with the smallest genotype effect value from the genome prediction model starting from the genotype with the smallest genotype effect value; evaluating the prediction accuracy of the model constructed after discarding the smaller genotypes; and selecting the corresponding genotype matrix to construct the genome prediction model according to the number of genotypes with the highest prediction accuracy of the genome prediction model. The present invention reduces the cost of genotyping and improves the prediction accuracy, so that the application range of genome prediction in human disease risk prediction and animal and plant breeding is more extensive.

Description

Technical Field

[0001] The present invention belongs to the technical fields of genomics, bioinformatics, genome prediction and genome breeding, and in particular relates to a genome prediction method, a prediction system and an application thereof. Background Art

[0002] At present, with the rapid development of high-throughput sequencing technology, researchers can accurately and quickly assemble and sequence the genome of target species, which has brought genetics and genomics research into a period of rapid development. High-quality genome assembly and cheap and efficient genotyping technology drive the acquisition of high-precision genotypes of research targets, thereby promoting the development of genome prediction. Genomic prediction is an efficient and powerful tool in the field of human complex traits, disease risk prediction and selective breeding. It uses phenotypic information and genotypic information to construct a genome prediction model in a reference population, which can predict the potential performance of a candidate population with only genotypic information.

[0003] The most classic genomic prediction model is the Genomic Best Linear Unbiased Prediction (GBLUP), which uses the genomic G matrix constructed by genotype information to replace the pedigree A matrix in the traditional best linear unbiased prediction, avoiding the error caused by Mendelian sampling, and thus achieving better prediction results. It is widely used in animal and plant selection breeding. In addition to the classic GBLUP, more genomic prediction models have been developed, such as various Bayesian methods. These genomic prediction models based on Bayesian methods have become the most widely used models in genomic prediction because of their higher prediction accuracy. However, due to the complex genetic regulatory mechanisms of different traits in different species, there is still no model that is optimal in all species and all traits, and the optimal model should still be selected according to specific species and traits.

[0004] In addition, widely used genomic prediction models use a large number of whole-genome genotypes and up to thousands of individuals to estimate the genomic breeding values ​​of candidate populations, which leads to serious overfitting in genomic prediction models and affects the prediction performance of the models. Although people have long been aware that too many genotypes will affect the prediction performance of the model, there is still no good way to solve this problem. In various machine learning, the prevention of overfitting has been widely mentioned, but there is almost no solution to the overfitting problem in genomic prediction models. Previous genomic prediction models optimized the prediction performance of the model by continuously optimizing the distribution of genotype effect values ​​and the distribution of effect value variances, but few have improved the prediction performance of genomic models from the perspective of preventing overfitting.

[0005] Based on the above factors, it is urgent to solve the overfitting problem caused by the introduction of high-dimensional genotype data in the genome prediction model, so as to reduce the noise interference of the model, improve the prediction performance of the genome prediction model, and make it more widely used in disease risk prediction and animal and plant breeding.

[0006] Through the above analysis, the problems and defects of the prior art are as follows:

[0007] (1) Traditional genomic prediction models use tens of thousands of genotypes to build prediction models, which leads to the "large p and small n" problem caused by the introduction of high-dimensional data, resulting in serious overfitting problems. Although existing models such as the Bayesian lasso optimize parameters and add penalty terms when building models, they still do not solve the fitting problem in genomic prediction, resulting in low genomic prediction accuracy in disease risk prediction and animal and plant breeding.

[0008] (2) In the process of constructing existing genome prediction models, it is necessary to genotype the variants on the entire genome, which makes the genotyping cost high. As a result, the application scope of genome prediction in disease risk prediction and animal and plant breeding is limited, and its practicality is poor.

[0009] The difficulty in solving the above problems and defects is: (1) the biological significance of the impact of genomic variation on phenotype is ignored, and the model is optimized purely from a data perspective without considering the burden of introducing genotype information unrelated to the trait on the model; (2) there is very little research on the overfitting phenomenon in the field of genomic prediction.

[0010] The significance of solving the above problems and defects is: to greatly improve the prediction accuracy of the existing genome prediction model, and to obtain high-precision prediction accuracy by typing very few genotypes related to the trait, thereby reducing the cost of genotyping. Summary of the invention

[0011] In order to overcome the problems existing in the related art, the disclosed embodiments of the present invention provide a genome prediction method, a prediction system, a readable storage medium and an application thereof, and specifically relate to a method for improving the prediction accuracy of a genome prediction model.

[0012] The technical solution is as follows: a genome prediction method, characterized in that it is applied to an information data processing terminal, and the genome prediction method includes: gradually discarding genotypes with relatively low effect values ​​in all genotype matrices, retaining only the optimal genotype matrix with a large effect value to construct a genome prediction model, and using genome information to predict the phenotypes of different species.

[0013] In one embodiment, the genome prediction method specifically comprises the following steps:

[0014] Step 1: measuring the phenotypic value of the target species trait to obtain phenotypic information, and performing genotyping on the individuals with the phenotypic value to obtain genotypic information;

[0015] Step 2: Use genotype information and phenotype information to build a genome prediction model based on all genotypes;

[0016] Step three, based on the correlation between phenotypic information and genotypic information, use maximum likelihood estimation or Bayesian method based on Gibbs sampling to estimate the effect value of each genotype in the constructed genomic prediction model (in the constructed model, based on different assumptions about the effect value and its variance distribution in different models, the corresponding parameters of the effect value distribution are obtained to estimate the effect value of each genotype. Specifically, based on the correlation between phenotypic information and genotypic information, use maximum likelihood estimation or Bayesian method based on Gibbs sampling in the constructed genomic prediction model to obtain the corresponding parameters of the distribution obeyed by the marker effect value, thereby estimating the effect value of each genotype).

[0017] Step 4, sorting all genotypes from small to large according to the size of each genotype effect value, and gradually discarding the genotype with the smallest genotype effect value from the genome prediction model starting from the genotype with the smallest genotype effect value; wherein, starting from the genotype with the smallest genotype effect value, a certain number (such as discarding 200 genotypes each time) are discarded from the genome prediction model, thereby constructing a new genome prediction model using the retained genotypes with larger effect values.

[0018] Step five, using the cross-validation method, by calculating the correlation between the genomic breeding values ​​and the actual phenotypic values ​​of the individuals with masked phenotypic values, the prediction accuracy of the genomic prediction model constructed by discarding the genotypes with smaller effect values ​​each time and retaining only the genotypes with larger effect values ​​is evaluated; wherein, by calculating the Pearson correlation coefficient or the area under the ROC curve of the genomic breeding values ​​and the actual phenotypic values ​​of the individuals with masked phenotypic values, the prediction accuracy of the genomic prediction model constructed by discarding the genotypes with smaller effect values ​​each time and retaining only the genotypes with larger effect values ​​is evaluated.

[0019] The calculation formula of Pearson correlation coefficient is:

[0020]

[0021] Where y is the masked phenotypic value, GEBV is the genomic breeding value; the area under the ROC curve is calculated as:

[0022]

[0023]

[0024] Where n1 is the number of surviving individuals, the phenotypic value is recorded as 1, n0 is the number of dead individuals, the phenotypic value is recorded as 0, and R1 is the sum of the rankings of the surviving individuals.

[0025] Step 6: After all genotypes are discarded from the genome prediction model, the prediction accuracy evaluation of the genome prediction model is terminated. According to the number of genotypes with the highest prediction accuracy of the genome prediction model, the corresponding genotype matrix is ​​selected to construct the genome prediction model, that is, the highest genome prediction data information is obtained.

[0026] The genome prediction model of the present invention is obtained by discarding markers with smaller effect values ​​and then iterating to obtain the optimal genotype matrix, which significantly reduces the overfitting problem caused by the introduction of high-dimensional data.

[0027] In one embodiment, the genotype information in step 1 is genotype information based on single nucleotide polymorphism typing, only the genotype with di-allelic sites is retained, and is encoded by 0, 1, and 2, representing the homozygous allele type AA, the heterozygous allele type Aa, and the minor homozygous allele type aa, respectively, and there is no missing genotype information used to construct the genome prediction model.

[0028] In one embodiment, the genome prediction model in step 2 is constructed by using all genotype data using an indirect method, and the constructed genome prediction models are all linear models based on additive effects;

[0029] The linear model formula is expressed as y = μ + Zu + e;

[0030] Where y is the phenotypic value of the individual, μ is the intercept, Z is the genotype matrix with n rows and m columns, where n is the number of samples used to build the genomic model, m is the number of genotypes, u is the effect value matrix with m rows and 1 column, and the i-th row corresponds to the effect value of the genotype in the i-th column of the Z matrix; e is the residual;

[0031] The sum of all genotypes multiplied by their effect values ​​is the genomic breeding value.

[0032] In one embodiment, the effect value estimation of the genotype in step three is performed in a genomic prediction model constructed from all genotype data, and the genomic prediction model generates an effect value for each genotype;

[0033] The effect value sorting in step 4 is performed from small to large according to the absolute value of each genotype effect value.

[0034] In one embodiment, the cross-validation method in step 5 includes: masking the phenotypic information of some individuals and retaining only the genotypic information, estimating the genomic breeding value of the masked phenotype individual using a genomic prediction model, and comparing the correlation with the actual phenotype to obtain prediction information of the genomic prediction model;

[0035] The genotype matrix used in step 6 is obtained by successively discarding genotypes with smaller effect values ​​and performing continuous iterations;

[0036] The genomic prediction model was constructed based on the optimal genotype matrix with the highest prediction accuracy determined by discarding markers with small effect sizes.

[0037] Another object of the present invention is to provide a genome prediction system including an information data processing terminal, the information data processing terminal including a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, the processor performs the following steps:

[0038] The genotypes with relatively low effect values ​​in the entire genotype matrix are gradually discarded, and only the optimal genotype matrix with large effect value is retained to construct a genome prediction model to predict the phenotypic and genotypic data of different species.

[0039] Another object of the present invention is to provide an application of the genome prediction method in predicting phenotypic and genotypic data of different species such as sorghum, dairy cow, horse, rainbow trout and carp.

[0040] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:

[0041] The present invention gradually discards genotypes with low effect values ​​in the genome prediction model and only retains genotypes with large effects, thereby reducing the overfitting damage caused by introducing high-dimensional genotype data into the model. Compared with the current common use of all genotype data to build a genome prediction model, the present invention significantly improves the accuracy of genome prediction, and only types genotypes with large effects can obtain the highest prediction accuracy, reducing the cost of genotyping, making the application range of genome prediction in human disease risk prediction and animal and plant breeding more extensive.

[0042] In the present invention, "all genotypes" refers to the prediction accuracy of the traditional genome prediction model constructed using all genotypes, and the "optimal genotype" refers to the prediction improvement of the present invention compared with the traditional genome prediction model constructed using all genotypes. Among the 12 traits of the five species tested, the present invention showed a significant improvement in prediction accuracy.

[0043] In addition, the present invention significantly improves the prediction accuracy of existing genomic prediction models, and only needs to type genotypes with large effects to meet the needs, and the prediction accuracy is significantly higher than the prediction model constructed using all genotypes, so that the prediction accuracy and applicability of large-scale sample typing have been significantly improved compared with previous prediction models.

[0044] The optimal genotype matrix obtained by the present invention can be used not only for various traditional genome prediction models based on additive effects, but also for various prediction models based on machine learning algorithms such as random forests and support vector machines.

[0045] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0047] Figure 1 It is a flow chart of the genome prediction method provided by an embodiment of the present invention.

[0048] Figure 2 is a graph showing the improvement in prediction accuracy of the optimal genotype matrix provided in an embodiment of the present invention relative to the genomic prediction model constructed using all genotype matrices in a sorghum data set; wherein, Figure 2(a) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix in sorghum stalk length trait relative to the genomic prediction model constructed using all genotype matrices; Figure 2(b) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix in sorghum stalk trait relative to the genomic prediction model constructed using all genotype matrices; Figure 2(c) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix in sorghum total weight trait relative to the genomic prediction model constructed using all genotype matrices; and Figure 2(d) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix in sorghum stalk number trait relative to the genomic prediction model constructed using all genotype matrices.

[0049] Figure 3 is a graph showing the improvement in prediction accuracy of the optimal genotype matrix provided by an embodiment of the present invention relative to the genomic prediction model constructed using all genotype matrices in the data sets of dairy cows and horses; wherein, Figure 3(a) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the milk fat rate trait of dairy cows; Figure 3(b) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the somatic cell score trait of dairy cows; Figure 3(c) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the milk production trait of dairy cows; and Figure 3(d) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the coat color trait of horses.

[0050] Figure 4 is a graph showing the improvement in prediction accuracy of the optimal genotype matrix provided by an embodiment of the present invention relative to the genomic prediction model constructed using all genotype matrices in the data sets of rainbow trout and carp; wherein, Figure 4(a) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the survival status traits of rainbow trout; Figure 4(b) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the survival time traits of rainbow trout; Figure 4(c) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the weight traits of carp; and Figure 4(d) is a graph showing the improvement in prediction accuracy of the optimal genotype matrix relative to the genomic prediction model constructed using all genotype matrices in the body length traits of carp. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific implementation disclosed below.

[0052] The present invention provides a method for solving the overfitting problem caused by the introduction of high-dimensional genotype data in the existing genome prediction model, that is, a genome prediction method, including: gradually discarding genotypes with low effect values, and only retaining the optimal genotype matrix with a large effect value for constructing a genome prediction model, so as to obtain the highest prediction accuracy. Compared with the current commonly used method of using all genotype information to construct a model and perform genome prediction, the present invention significantly improves the prediction performance of the genome prediction model, and only typing the optimal genotype matrix can obtain the highest prediction accuracy, thereby reducing the cost of genotyping.

[0053] The technical solution of the present invention is further described below in conjunction with specific embodiments.

[0054] Embodiment 1:

[0055] like Figure 1 As shown, the present invention provides a genome prediction method comprising:

[0056] S101, measure the phenotypic value of the target trait under study to obtain phenotypic information (such as growth, weight, disease risk, etc.), and perform genotyping on individuals with phenotypic values ​​(genotyping can be performed using whole genome resequencing, simplified genome sequencing, gene chips, etc.) to obtain genotypic information.

[0057] S102, using genotype information and phenotype information to construct a genome prediction model based on all genotypes.

[0058] S103, based on the correlation between phenotypic information and genotypic information, the effect value of each genotype in the constructed genomic prediction model is estimated using maximum likelihood estimation or a Bayesian method based on Gibbs sampling.

[0059] S104, sorting all genotypes from small to large according to the size of each genotype effect value, and gradually discarding the genotype with the smallest genotype effect value from the genome prediction model.

[0060] S105, using the cross-validation method, the prediction accuracy of the model is evaluated after discarding the genotype with a smaller effect value each time by calculating the correlation between the genomic breeding value and the actual phenotypic value of the individuals with masked phenotypic values.

[0061] S106, after all genotypes are discarded from the genome prediction model, the prediction accuracy evaluation of the model is ended, and the corresponding genotype matrix is ​​selected to construct the genome prediction model according to the number of genotypes that appear in the model with the highest prediction accuracy, so as to obtain the highest genome prediction accuracy.

[0062] In a preferred embodiment of the present invention, the genotype information of step S101 is genotype information based on single nucleotide polymorphism typing, only the genotype with di-allelic sites is retained, and is encoded by 0, 1, and 2, representing the major homozygous allele type AA, the heterozygous allele type Aa, and the minor homozygous allele type aa, respectively, and the genotype information used to construct the genome prediction model cannot be missing.

[0063] In a preferred embodiment of the present invention, the genome prediction model construction in step S102 is constructed using an indirect method using all genotype data, and the constructed genome prediction models are all linear models based on additive effects;

[0064] The linear model formula is expressed as y = μ + Zu + e;

[0065] Where y is the phenotypic value of the individual, μ is the intercept, Z is the genotype matrix with n rows and m columns, where n is the number of samples used to build the genomic model, m is the number of genotypes, u is the effect value matrix with m rows and 1 column, the i-th row corresponds to the effect value of the genotype in the i-th column of the Z matrix; e is the residual;

[0066] The sum of the products of all genotypes and their effect values ​​is the genomic breeding value.

[0067] In a preferred embodiment of the present invention, the step S103 obtains corresponding parameters of the effect value distribution in the constructed model based on different assumptions of the effect value and its variance distribution in different models, thereby estimating the effect value of each genotype.

[0068] In a preferred embodiment of the present invention, the effect value estimation in step S103 is performed in a genome prediction model constructed from all genotype data, and the model generates an effect value for each genotype.

[0069] In a preferred embodiment of the present invention, the effect value sorting in step S104 is performed based on the absolute value of each genotype effect value from small to large, rather than the numerical value.

[0070] In a preferred embodiment of the present invention, the prediction accuracy of the genomic prediction model in step S105 is calculated by a cross-validation method, by masking the phenotypic information of some individuals, retaining only the genotypic information, estimating the genomic breeding value of the masked phenotype individual using the genomic prediction model, and comparing the correlation with the actual phenotype, thereby obtaining the prediction accuracy of the model. The metric for the prediction accuracy of binary traits is the area under the ROC curve of the genomic breeding value and the actual phenotypic value, and the metric for the prediction accuracy of non-binary traits is the Pearson correlation coefficient of the genomic breeding value and the actual phenotypic value, and the closer the value is to 1, the higher the prediction accuracy.

[0071] The calculation formula of Pearson correlation coefficient is:

[0072]

[0073] Where y is the masked phenotypic value, GEBV is the genomic breeding value; the area under the ROC curve is calculated as:

[0074]

[0075]

[0076] Where n1 is the number of surviving individuals, the phenotypic value is recorded as 1, n0 is the number of dead individuals, the phenotypic value is recorded as 0, and R1 is the sum of the rankings of the surviving individuals.

[0077] In a preferred embodiment of the present invention, the genotype matrix used by the genome prediction model with the highest prediction accuracy is obtained by successively discarding genotypes with smaller effect values ​​through continuous iteration.

[0078] In a preferred embodiment of the present invention, the prediction accuracy of the model constructed after discarding the genotype with a smaller effect value each time is evaluated by calculating the Pearson correlation coefficient or the area under the ROC curve of the genomic breeding value and the actual phenotypic value of the individual with masked phenotypic value.

[0079] In a preferred embodiment of the present invention, the model ultimately used for genome prediction in step S106 is constructed based on the optimal genotype matrix with the highest prediction accuracy determined by discarding markers with smaller effect values, rather than the currently commonly used genome prediction model constructed using all genotypes.

[0080] Embodiment 2:

[0081] The present invention provides a genome prediction method (a method for improving the prediction accuracy of a genome prediction model), comprising the following steps:

[0082] (1) Collection of phenotypic and genotypic data for different species, including sorghum, dairy cow, horse, rainbow trout, and carp.

[0083] The genotype data of sorghum included 3260 genotypes and 1020 samples, which were genotyped by restriction enzyme site-associated DNA sequencing. The phenotypic data were culm length (CL), total weight (TW), culm diameter (CD), and culm number (CN). The data of cattle included 42551 genotypes and 5024 samples, which were genotyped by 50KSNP chip. The phenotypic data were milk fat percentage (fp), somatic cell score (scs), and milk production (my). The data of horses included 50621 genotypes and 480 samples, which were genotyped by 54KSNP chip. The phenotypic data were coat color (cc). The data of rainbow trout included 1934 samples and 27490 genotypes, which were genotyped by 57KSNP chip. The phenotypic data were survival time (sd) and survival status (ss) after salmon Rickettsia challenge. Survival status was a binary trait representing whether each individual died or survived after the challenge test, recorded as 0 or 1. The data for common carp included 1259 samples and 15615 genotypes, genotyping was performed by restriction site-associated DNA sequencing, and the phenotypes were standard length (sl) and body weight (bw).The above data were obtained from published articles, including sorghum (Ishimori M, Takanashi H, Hamazaki K et al. Dissecting the Genetic Architecture of Biofuel-Related Traits in a Sorghum Breeding Population. G3 Genes | Genomes | Genetics 2020; 10: 4565-4577), dairy cow (Accuracy of whole-genome prediction using a genetic architecture-enhanced variance-covariance matrix. G3 Genes | Genomes | Genetics 2015; 5: 615-627), horse (KAML: improving genomic prediction accuracy of complex traits using machine learning determined parameters. Genome Biology 2020; 21: 146), rainbow trout (Yoshida GM, Bangera R, Carvalheiro R et al. Genomic prediction accuracy for resistance against Piscirickettsia salmonis in farmed rainbow trout. G3 Genes|Genomes|Genetics 2018; 8:719-726), carp (Palaiokostas C, Vesely T, Kocour M et al. Optimizing genomic prediction of host resistance to koi herpesvirus disease in carp. Frontiers in Genetics 2019; 10:543).

[0084] (2) Construction of genome prediction model

[0085] All genotype and phenotype data were imported into R, and a genomic prediction model was constructed. The genomic prediction model was constructed using the R packages BGLR and rrBLUP, which correspond to the genomic prediction models constructed by the Bayesian method and the ridge regression best linear unbiased prediction method, respectively.

[0086] (3) Estimation and ranking of genotype effect values

[0087] The constructed genomic prediction model was used to generate effect values ​​for each genotype, and then the effect values ​​were sorted according to their absolute values. The rrBLUP method used maximum likelihood estimation to generate effect values ​​for each genotype, while various Bayesian methods constructed Markov chain Monte Carlo chains through Gibbs sampling, sampling from the posterior distribution of each genotype parameter to calculate the effect value of each genotype. According to the size of the effect value, the order function in R was used to sort the genotypes in order from small to large genotype effect values.

[0088] (4) Determination of the genotype matrix with the highest prediction accuracy

[0089] According to the ranking size of the genotype effect value, it is gradually discarded from the genomic prediction model, and the number of discards each time is set to 200 genotypes. After discarding 200 genotypes, the genomic prediction model constructed by the existing genotype matrix is ​​subjected to five-fold cross validation to evaluate its prediction accuracy until the genomic prediction model no longer contains genotype data. The specific details of the implementation of the five-fold cross validation are as follows: all genotype data and phenotypic data are randomly divided into five subsets, one of which has phenotypic information masked and only retains genotypic information. The other four subsets are used to construct the genomic prediction model and calculate the effect value of the genotype. The genotype effect value matrix is ​​matrix multiplied with the genotype data of the masked phenotype subset to generate a genomic breeding value for each individual. If the phenotypic data is a continuous trait, the Pearson correlation coefficient between the genomic breeding value and the phenotype is used to represent the accuracy. If it is a binary trait, the area under the ROC curve is used as the accuracy. In order to prevent sampling errors during the implementation of the five-fold cross validation, all prediction accuracy evaluations are performed 50 times, and the mean and standard deviation of the 50 repetitions are used as the final results.

[0090] (5) In order to verify the effectiveness of the present invention, all the data sets in (1) were used to verify the present invention. Figure 2 shows the improvement of the present invention in the four phenotypic traits of sorghum. Compared with the genomic prediction model constructed using all genotype matrices, the accuracy of the model constructed using the optimal genotype matrix is ​​significantly improved, and it is applicable to all four traits. Figure 3 shows the improvement in the prediction accuracy of the genomic prediction model in the cattle and horse data sets, and Figure 4 shows the improvement in the prediction accuracy of the genomic prediction model for rainbow trout and carp. The 12 traits of the five species tested all showed that the optimal genomic prediction model constructed based on discarding small-effect genotypes had higher prediction accuracy than the genomic prediction model constructed using all genotypes, indicating that this method is applicable to different traits and different species.

[0091] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure disclosed herein. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

[0092] It should be understood that the present disclosure is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure should be limited by the appended claims.

Claims

1. A genome prediction method, characterized in that: The genome prediction method specifically comprises the following steps: Step 1: measuring the phenotypic value of the target species trait to obtain phenotypic information, and performing genotyping on the individuals with the phenotypic value to obtain genotypic information; Step 2: Use genotype information and phenotype information to build a genome prediction model based on all genotypes; Step 3: Based on the correlation between phenotypic information and genotypic information, the effect value of each genotype in the constructed genomic prediction model is estimated using maximum likelihood estimation or Bayesian method based on Gibbs sampling; Step 4, sorting all genotypes from small to large according to the size of each genotype effect value, and gradually discarding the genotype with the smallest genotype effect value from the genome prediction model starting from the genotype with the smallest genotype effect value; Step 5: Using the cross-validation method, the prediction accuracy of the model is evaluated after discarding the genotype with a small effect value each time by calculating the correlation between the genomic breeding value and the actual phenotypic value of the individuals with masked phenotypic values; Step 6: After all genotypes are discarded from the genome prediction model, the prediction accuracy evaluation of the genome prediction model is terminated, and the corresponding genotype matrix is ​​selected according to the number of genotypes with the highest prediction accuracy of the genome prediction model to construct the genome prediction model and obtain the highest genome prediction accuracy; In step 2, the genomic prediction model was constructed using the indirect method using all genotype data, and the constructed genomic prediction models were all linear models based on additive effects; The linear model formula is expressed as y=μ+Zu+e; Where y is the phenotypic value of the individual, μ is the intercept, Z is the genotype matrix with n rows and m columns, where n is the number of samples used to build the genomic model, m is the number of genotypes, u is the effect value matrix with m rows and 1 column, and the i-th row corresponds to the effect value of the genotype in the i-th column of the Z matrix; e is the residual; The sum of all genotypes multiplied by their effect values ​​is the genomic breeding value.

2. The genome prediction method according to claim 1, characterized in that: In step one, the genotype information is the genotype information based on single nucleotide polymorphism typing, only the genotype with di-allelic sites is retained and encoded by 0, 1, and 2, representing the homozygous allele type AA, the heterozygous allele type Aa, and the minor homozygous allele type aa, respectively, and there is no missing genotype information used to construct the genome prediction model.

3. The genome prediction method according to claim 1, characterized in that: In step three, the effect value estimation of genotypes is performed in the genomic prediction model constructed from all genotype data, which generates effect values ​​for each genotype.

4. The genome prediction method according to claim 1, characterized in that: The effect value sorting in step 4 is performed from small to large according to the absolute value of each genotype effect value.

5. The genome prediction method according to claim 1, characterized in that: The cross-validation method in step 5 includes: masking the phenotypic information of some individuals and retaining only the genotypic information, estimating the genomic breeding value of the masked phenotype individual using a genomic prediction model, and comparing the correlation with the actual phenotype to obtain the prediction information of the genomic prediction model; The calculation of the correlation between the genomic breeding value and the actual phenotypic value of the individual with the masked phenotypic value so as to evaluate the prediction accuracy of the model constructed by discarding the genotype with smaller effect value each time and retaining only the genotype with larger effect value comprises: The prediction accuracy of the genomic prediction model constructed by discarding genotypes with smaller effect values ​​and retaining only genotypes with larger effect values ​​was evaluated by calculating the Pearson correlation coefficient or the area under the ROC curve between the genomic breeding value and the actual phenotypic value of the individuals with masked phenotypic values; The calculation formula of Pearson correlation coefficient is: ； Where y is the masked phenotypic value, GEBV is the genomic breeding value; the area under the ROC curve is calculated as: ； ； Where n1 is the number of surviving individuals, the phenotypic value is recorded as 1, n0 is the number of dead individuals, the phenotypic value is recorded as 0, and R1 is the sum of the rankings of the surviving individuals.

6. The genome prediction method according to claim 1, characterized in that: The genotype matrix used in step 6 is obtained by successively discarding genotypes with smaller effect values ​​and performing continuous iterations; The genomic prediction model was constructed based on the optimal genotype matrix with the highest prediction accuracy determined by discarding markers with small effect sizes.

7. A genome prediction system for implementing the genome prediction method according to any one of claims 1 to 6, characterized in that: The genome prediction system includes an information data processing terminal, the information data processing terminal includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps: The genotypes with relatively low effect values ​​in the entire genotype matrix are gradually discarded, and only the optimal genotype matrix with large effect values ​​is retained to construct a genomic prediction model, and the phenotypes of different species are predicted using genomic information.

8. An application of the genome prediction method according to any one of claims 1 to 6 in predicting phenotypic and genotypic data of different species such as sorghum, dairy cow, horse, rainbow trout and carp.

Images