For example, Habier et al. [ 1] showed that, in the German Holstein breed, accuracies of genomic predictions of animals that were distantly-related to the reference population declined.
However, the standard errors of our results do not allow us to draw an unambiguous conclusion on whether the latter three estimates are significantly different from 0. For the Brahman breed, accuracies published by [ 7] for g.WW, g.YW and g.SEMA were also considerably higher than those obtained from ALL PE.
Further improvements of across-breed accuracy of genomic predictions and QTL mapping could be achieved by increasing the size of the reference population, including more breeds, and possibly by exploiting pleiotropic effects to improve mapping efficiency for QTL with small effects.
However, Hayes et al. [ 23] showed that, in sheep breeds, accuracy of imputation in single-breed analyses was higher than accuracy of imputation in a multi-breed analysis.
Toosi et al. [ 9] reported that using training and validation populations from the same breed resulted in the highest values of accuracy of GEBV in other cross or admixed populations and that compared to these values, when training and validation were done in different breeds, accuracy dropped by 46%.
Averaged across traits within breeds, accuracies of GEBV were highest when PE from the pooled across-breed training population were used.
Similar poor between-breed accuracies were reported by Daetwyler et al. [ 4] who also showed that most of the observed accuracy was due to population structure.
As in scenarios with equal allele substitution effects across breeds, accuracies obtained with the formula based on selection index theory were in good agreement with empirical accuracies.
Besides differences in LD across breeds, the accuracy of multi-breed genomic prediction might also be influenced by other factors, such as the absence of family relationships or differences in allele frequencies, e.g. [ 2, 13, 49].
In all breeds, imputation accuracy was higher than 97%, which indicates that imputation to the high-density chip was accurate.
Within each breed, realized accuracies were estimated using the correlation from the GenSel analysis and heritabilities estimated from the corresponding within-breed GBLUP analysis using procedures outlined in McClure et al. [ 20].
