Comparison Of Selection Efficiency In Several Genomic Evaluating Methods

From Meuwissen come up with using Bayes methods for whole genome selection and prove the feasibility of whole genome selection in animals in2001. This method became the hot spot in whole genome selection because its high accuracy, efficiency and advancement.The aim of this article was to study several methods in whole genome selection (Bayesian regression method, GBLUP and SVM). First by derivating those methods mentioned above, some bugs of Meuwissen’s formula in Bayes A were found and revised in this paper. And then encode the genotype data and rewrite some program of the corresponding those methods based on the1000×50k SNP data of phenotype genotype analyzed by QMSim. Those rewrite program first calculate the correlation coefficient of predictive value and phenotype value and compared MSE and operation time, and then using ten-times cross validation by simulated pig chip data, obtained the optimal method for pig whole genome selection in the future. The specific results were described as blow:Correlation coefficient of those data arranged form from high to low were: SVM(0.9214), Bayes Cpi(0.8848),Bayes A(0.8678), Bayes Ridge(0.848),Bayes LASSO(0.8449), GBLUP (0.8374). MSE were:GBLUP(1.1035)、Bayes LASSO(1.0689)、Bayes Ridge(1.0523)、Bayes A(0.9124)、Bayes Cpi(0.7336)、 SVM(0.2953), respectively. And the run time were SVM(13.45min)、Bayes Ridge(53.98min)、GBLUP(1.02h)、Bayes LASSO(1.53h)、Bayes A(9.59d)、Bayes Cpi(11.51d).Results of cross validation showed that the correlation coefficient of SVM、 Bayes A、Bayes Cpi is better than the rest (SVM0.919、Bayes Cpi0.8749、Bayes A0.8682). SVM was the fastest method (run time:2.93h). The run time of Bayes Cpi and Bayes A were too slow (11.29d and9.41d) if calculate only once cross validation.In conclusion, machine learning algorithm and SVM is better than other methods not only in run time but also in correlation coefficient. Thus those can be the mainly methods in pig whole genome selection. Bayes A and Bayes Cpi algorithm have an advantage is correlation coefficient but short in run time. If solved the problem of the operation time, Bayes A and Bayes Cpi algorithm still can be used as all the major genetic selection.Considering run time and correlation coefficient,Bayes LASSO and Bayes Ridge were the mainly Bayes method in pig whole genome selection.