The Algorithm Research For Genomic Selection Study Based On Machine Learning

Genomic selection(GS)has been widely used in animal and plant breeding,in which genomic best linear unbiased prediction(GBLUP)based on linear mixed models and Bayesian methods(Bayes A,Bayes B,Bayes C,Bayes C?)based on prior information is mainly applied in animal and plant breeding.However,these algorithms naively assumed additive inheritance,ignored the interaction between genes and nonlinear prediction models.Machine learning(ML)improves the predictive performance of the model by learning "experience" from data.It does not need to establish assumptions and can construct a nonlinear predictive model.To further improved the prediction of GS,this study applied ML to construct a non-linear model to estimate the genomic estimated breeding value(GEBV)and fully mined the genomic information.(1)This study evaluated the prediction accuracy of Support vector regression(SVR),Random forest(RF),Kernel ridge regression(KRR)and Boosting integration algorithms in the GEBV estimation of important economic traits of Chinese Simmental cattle(carcass weight,live weight before slaughter,eye muscle area).The results found that in the genomic prediction of carcass weight,live weight and eye muscle area traits of Chinese meat Simmental cattle,the GEBV prediction accuracy of the SVR,RF,KRR and Adaboost.RT was higher than the GBLUP method.The KRR method possessed the highest accuracy,an average increase of 14.8% compared to the GBLUP method.(2)This study had developed a Stacking ensemble learning framework(SELF)integrating SVR,KRR,and ENET(Elastic net)three machine learning methods to apply to GS,and validated using three kinds of animal and plant data sets,including Chinese meat Simmental cattle.The results showed that the prediction of SELF was higher than that of GBLUP and Bayes B methods.Compared with SVR,ENET,KRR and GBLUP methods,the average degree of improvement of the prediction accuracy of SELF was8.42%,6.03%,5.31% and 7.71% respectively.This study once again proved that machine learning had great application potential in genomic selection.It had a huge advantage in the prediction of GEBV of Chinese meat Simmental cattle population,and further improved the accuracy of genome prediction.At the same time,SELF provided a brand-new and efficient machine learning algorithm for GS,which further enriched and expanded the method system of GS.