Use of Principal Component and Factor Analysis to reduce the number of independent variables in the prediction of Genomic Breeding Values

Abstract

On a simulated population of 2,500 individuals, Principal Component Analysis and Factor Analysis were used to reduce the number of independent variables for the prediction of GEBVs. A genome of 100 cM with 300 bialleic SNPs and 20 multiallelic QTLs was considered. Two heritabilities (0.2 and 0.5) were tested. Multivariate reduction methods performed better than the traditional BLUP with all the SNPs, either on generations with phenotypes available or on those without phenotypes, especially in the low heritability scenario (about 0.70 vs. 0.45 in generations without phenotypes). The use of multivariate reduction techniques on the considered data set resulted in a simplification of calculations (reduction of about 90% of predictors) and in an improvement of GEBV accuracies.