Genomic Selection Models Of Crop Hybrids’ Phenotype

Genomic selection(GS)breeding technology can evaluate the breeding values based on individual genotypes,which can greatly shorten the breeding cycle.GS has become a new technology in animal and plant breeding.Especially in the cross breeding of crops,GS has a more prominent advantage,because the genotype of hybrids can be inferred from the parental genotype,and untested hybrids can be predicted only by obtaining the genotype of the parents and phenotype of a few hybrids.Accurate genomic prediction is a prerequisite for the success of GS.However,the accuracy of genomic prediction is often low for some complex traits,especially for yield traits that are highly influenced by environment.In order to further improve the accuracy of phenotypic prediction of rice hybrids,we proposed a new strategy of incorporating parental phenotypic information into rice hybrids prediction and identify the optimal model of this strategy.At the same time,we developed a new method of hybrid maize phenotypic prediction based on auxiliary traits,so as to achieve more comprehensive and reliable prediction and selection.The results provided important theoretical and technical support for precision breeding of rice and maize.The research includes the following two parts:1.Genomic selection method for predicting yield-related traits in rice hybrids by integrating parental phenotypesBased on 210 recombination inbred lines of rice genome information and phenotypic data of 278 hybrids,we discussed incorporating parental phenotype will improve the predictive accuracy of yield-related traits of hybrid,and compared 13 GS methods using cross validation.The 13 GS methods include parametric,semi-parametric and nonparametric method of prediction accuracy.To investigate the effects of different methods on the accuracy of phenotypic prediction of different traits.Parametric methods includes:Genomic Best Linear Unbiased Prediction(GBLUP),Least absolute shrinkage and selection operator(LASSO),BayesA,BayesB,BayesC,Partial Least Squares(PLS),Elastic Net(EN),Ridge Regression(RR);Semi-parametric methods are as follows:Reproducing Kernel Hilbert Space(RKHS);Non-parametric methods include Support Vector Machine(SVM)and Random Forest(RF).The results showed that the combination of parental phenotypes could significantly improve the accuracy of hybrid prediction,and there are significant differences in the predictability of different methods.In general,among the 13 methods,the predictability of GBLUP and BayesB are better,while the predictability of support vector machine and partial least squares method was poor.In addition,there were significant differences in the predictability of different traits.Among the four traits,the 1000-grain weight had the highest predictability,and the yield had the lowest predictability,indicating that the higher the heritability,the higher the predictability.Combined with parental phenotypes,the predictability of yield,tiller number,grain number and 1000-grain weight increased by 12.5%,24%,7.4%and 5.8%,respectively.This study is beneficial to improve the prediction accuracy of yield-related traits,and lays a solid theoretical foundation for precision breeding of rice hybrid phenotypes.2.Genomic selection strategy of maize hybrids with auxiliary traitsThere is usually a certain correlation between biological traits,and multi-trait joint analysis can effectively improve the accuracy of prediction by using genetic correlation or environmental information among traits.In this study,the phenotypic values of target traits were predicted using known or predicted auxiliary traits related to target traits,and a new strategy of hybrid GS based on auxiliary traits was discussed.Using maize data set,846 hybrids from 257 inbred lines with traits including plant height,ear height,ear weight and grain weight.Genotype and phenotypic data of parental materials were obtained,and genomic prediction of maize hybrids was performed by combining two strategies of known auxiliary traits and predicted auxiliary traits.The results showed that genomic prediction combined with known auxiliary traits could significantly improve the predictability of target traits.Using different combinations of auxiliary traits,the predictability of plant height,ear height,ear weight and grain weight increased by 20.5%,32.0%,50.2%and 81.2%on average.Compared with the prediction of the known auxiliary traits,the combined predicted auxiliary traits had a smaller prediction advantage,but the predictability was still improved.The predictability of the four traits,plant height,ear height,ear weight and grain weight,increased by 1%,1.4%,3.4%and 4.3%,respectively.In addition,the phenotypic values of all possible hybrid combinations were predicted based on the identified hybrids,and the mean ear weight of the top 100 hybrids was 17.8%higher than the mean ear weight of all hybrids.There were extremely significant differences between the predicted Top100 and Bottom 100 cross combinations.This study provides a new idea for hybrid breeding of maize and other crops by GS technology.