Optimization of Training Set in Genome Wide Association Study and Genomic Evaluation

The optimization of the reference population in genomic evaluation plays an important role in livestock breeding, because of its potential impact on the accuracy of estimating the marker effects and genomic breeding values. In the present study, seven different train set selection methods including selection of all, selection of the highest and lowest performances, random selection, selection of individuals with the most and least marker and QTL similarity were evaluated. In genome wide association study selection of all as train set detected common SNPs which make a high variation on the trait. However selective train set was just reported rare SNPs with a major effect on the trait. In genomic selection simultaneous use of high-density markers and selective train set in comparison with low-density and selection of all as train set reduced accuracy, but did not change the ranking of animals. There was also an interaction between train set selection method and generation (P≤0.0134) as well as the linkage disequilibrium (P≤ 2e-16). In general, selection of all animals as a train set resulted in higher accuracy compared to six selective train set methods. There were no differences between the methods of selecting train set in populations with a low effective size (r2 = 0.255, Ne =100), but in populations with a high effective size (r2 = 0.086, Ne =400) methods, with different accuracy predicted genomic breeding values. The highest and lowest accuracy were respectively belonged to most QTL and marker similarity methods.