Predicting the accuracy of multi-population genomic prediction

Wientjes, Y.C.J.; Bijma, P.; Veerkamp, R.F.; Calus, M.P.L.


For designing breeding programs, knowing the accuracy of genomic prediction of different designs is essential for maximising response to selection. Accuracy of genomic prediction when using a single reference population can be predicted by the deterministic equation derived by Daetwyler et al. Our aim was to derive a deterministic equation to predict the accuracy when multiple populations are used for genomic prediction. The equation was derived based on a selection index approach for a scenario with population A and B in the reference population (RP), and using population C as selection candidates (SC). It was assumed that ng independent quantitative trait loci (QTL) were affecting the trait, that were the same across populations, and each QTL explained the same part of the genetic variance. The derived equation contains population parameters, such as ng, heritability and size of the populations in RP, and genetic correlations between populations. Due to linkage disequilibrium, however, loci are not independent. Therefore, ng was replaced by the number of effective chromosome segments, as was described for the equation for single population scenarios. The derived equation can also be applied in situations with SC from population A or B included in RP and in situations with more than two populations in RP. The equation was validated using a GBLUP model with simulated phenotypes and real genotypes of one Holstein Friesian population. This population was split in two populations for RP (population A and B) and one population for SC (population C), using different values for heritability and genetic correlations between the populations. Three different cattle breeds with real genotypes and simulated phenotypes, assuming different genetic correlations, were also used for validation.