Multi-Population Genomic Selection in Pigs

A large number of genotyped and phenotyped animals are needed in a reference dataset to train  models for obtaining high accurate genomic breeding values (GBV). The need for a large reference set can limit the application of genomic selection in pigs because of genotyping costs as well as small population sizes. These issues can be overcome by combining animals of different lines into one reference set (multi-population). Here, the accuracy of GS in pigs using multi-population reference sets was evaluated. Data for this study consisted of back-fat thickness phenotypes and genotypes (Porcine SNP60 Beadchip) of two sire lines (SL1, n=1,146; SL2, n=682). The number of animals of one population in the reference set was constant (600 individuals) and animals from SL2 were added in steps of 100, at random or selecting those most related to SL1, until a reference set of 1200 animals was reached. Phenotypes were adjusted for fixed effects of sex, herd-year-month and the covariate weight and subsequently used to obtain GBV from a Genomic BLUP model. Accuracy of the GBV was computed as the Pearson correlation between the GBV and the adjusted phenotype divided by the square root of the heritability. Using only animals from SL1 the estimated accuracy was 0.8. When animals from SL2 were added the accuracy increased and reached values between 0.89 - 0.91 for a reference set of 1200 animals. The increase in accuracies was similar when the animals were added at random or based on their relationship to SL1.