Effects of Selective Genotyping and Selective Imputation in Single-Step GBLUP

Abstract

Single-Step genomic prediction has been advocated to be the next logical step in the development of large routine applications. Despite its intriguingly simple concept, there are still several problems to be solved from a theoretical standpoint. One problem is the inflation of genomic predictions frequently occurring in practical applications. Ad-hoc remedies have been proposed, like the use of scaling-factors when building the H matrix or pruning of data and pedigrees used for prediction. In this investigation we suppose that selective genotyping and selective imputation in Single-Step GBLUP are major components contributing to inflated predictions. Using the general reformulation of the Single-Step model given by R. Fernando as an illustration, single-step GBLUP can be conceptually divided into two separate steps of estimation: step one is the estimation of gene-contents for all animals in the pedigree from observed genotypes and step two is the estimation of SNP-effects using observed and imputed genotypes and the available phenotypic data. In recent studies we have examined the effect of selective genotyping and the role of genotypes without phenotypes in Single-Step GBLUP in simulation studies. Our conclusion is that selective genotyping and the selective quality of genotype imputation can lead to biased genomic estimates in Single-Step GBLUP. In order to support our argument we demonstrate the effect of the inclusion or exclusion of older birth years of genotyped bulls showing strong evidence for selective genotyping on two-step and single-step predictions using real Fleckvieh data and discuss the vital importance of the difference between exclusion of genotypes and exclusion of phenotypes and genotypes.