Accounting for population structure in genomic prediction of Australian merino sheep

Abstract

The aim of this study was to compare different ways of accounting for population structure for genomic prediction of three economic traits in an Australian Merino sheep population. Population structure was accounted for either by fitting genetic groups (GG) derived from pedigree, or fitting principal components (PCs) calculated from the genomic relationship matrix based on 50k density SNP marker genotypes. Genomic breeding values (GBV) were calculated using genomic best linear unbiased prediction (GBLUP) and the GBV accuracy was evaluated based on 5 fold cross-validation across half-sib families. Best linear unbiased estimation (BLUE) of GG or PC effects were added to the GBV. Results showed that accounting for population structure either by fitting GG or PCs improved the accuracy of genomic prediction. Furthermore, fitting the first two PCs gave a similar accuracy to fitting GG derived from pedigree. The improvement in GBV accuracy after accounting for population structure in studied traits was not high (3.8% when averaged across traits) which may be because the genomic relationship matrix will implicitly account for some of the population structure effect when the GG or PCs are not fitted in analysis. In the case of missing or incomplete pedigrees, PCs can be used to account for population structure and to improve the prediction accuracies.