Including genetic groups as fixed or random effects in large scale single-step genomic predictions

Abstract

Genetic groups are often used in genetic evaluations to account for selection that cannot be accounted for by kinship. The genetic groups can be fitted as random or fixed effects, both in traditional BLUP and in single-step genomic predictions (ssGBLUP). In this study, we investigate how inclusion of genetic groups affect the predictive ability and bias in a large-scale three-trait ssGBLUP model for kg of milk, fat percentage, and protein percentage in Norwegian Red cattle. Including genetic groups as random effects rather than fixed effects reduced bias of breeding values. Specifically, for genotyped animals and ungenotyped animals with genotyped offspring and with sparse pedigree data, the bias of the Mendelian term of breeding values was reduced by about 25%, when fitting genetic groups as random compared to fixed.