Use Of Genomic Selection And Breeding Simulation In Parental Selection And Cross Prediction In Pure-Line Crop Breeding

Genomic selection(GS)is an emergent molecular breeding technology that utilizes genotype and phenotype in training population for statistiscal modeling,then predicts trait values in breeding population for selection.At present,many prediction models have been developed to predict genomic estimated breeding value(GEBV),such as ridge regression best linear unbiased prediction(RR-BLUP),genomic best linear unbiased prediction(GBLUP),Bayes model and a few machine learning models.Current GS studies are mostly focused on the prediction of traits themselves,but there is little research on parental selection,which is an important step in breeding,especially in pure line breeding crops.In addition,epistasis was often ingored in prediction for simplicity.Genotypic by environment interaction has always been an important aspect of crop breeding.By using the genetic correlation of traits in different environments,the performance of traits in specific environments can be predicted.The main purpose of this study is to explore the use of GS and breeding simulation methods to carry out parental selection and cross prediction of wheat(Triticum aestivum L.),to compare the predictability of different GS models for modeling epistatic effects on different populations and traits,and to compare single environmental prediction and multi-environmental prediction of different GS models.The main research contents and results obtained are as follows: 1.Use genomic selection and breeding simulation for cross predictionUnder the different genetic structure of traits,the difference of different genomic selection models on cross prediction were investigated by simulation.The genetic gains of predicted crosses with different selection intensities were compared.The results showed that the prediction accuracy of different models had no significant difference under different genetic structure of traits.Under all set selection intensities,the genetic gains obtained by crossing prediction through the usefulness were higher than those obtained by selected through mid-parent value.2.Study on parental selection for wheat yield and quality improvementThe performance of all possible crosses in a high-quality wheat parent population were predicted by predicting the usefulness of cross.The genetic gain and genetic diversity of the selected progeny in yield and quality traits were compared under four different parental selection schemes.The results showed that the selection index including both yield and quality traits could effectively obtain genetic gain for both yield and quality traits,but also retain more genetic diversity which is conducive to obtaining continuous genetic gain in long-term selection.3.Study on fitting epistasis effect in genomic selectionUsing two wheat natural populations and one rice(Oryza sativa)recombinant inbred lines(RIL)population,the GS models fitting epistasis effect were studied and compared with the GS model fitting only additive effect.The results showed that adding the epistasis effect in GS model will improve the prediction accuracy in most cases.In a few traits and environments,the model with epistasis effect is equivalent to or slightly lower than that with only additive effect.Therefore,the epistasis effect should be considered as far as possible in the prediction model when GS method is used in the selection of pure line rice and wheat varieties.4.Multiple environmental phenotypes prediction in genomic selectionA multi environmental GS study was conducted using a multi location experiment of a rice RIL population.The cross validation schemes of two breeding scenarios were used to compare the prediction accuracy of different models.The results showed that when the predicted varieties have no observed values in all environments,the multi-environment prediction model behaved similarly to the single-environment prediction model;when the predicted varieties have observed values in other environments,the prediction accuracy of the multi-environment model was much higher than that of the single-environment model.Therefore,multi-environment model can effectively utilize the correlation between two environments,thus improving the prediction accuracy.