Fig. 9

Potential accuracy of predicting the genomic value of individuals from population \(\varvec{A}\) under different models: within-population, single \({\mathbf{GRM}}\) (WPSG), within-population, multiple \({\mathbf{GRM}}\) (WPMG), multi-population, single \({\mathbf{GRM}}\) (MPSG), multi-population, multiple \({\mathbf{GRM}}\) (MPMG), in relation to different levels of underestimation of \({\mathbf{M}}_{{\mathbf{e}}}\) within the predicted population \(\varvec{A}\) (0 – 90%). The following assumptions were made: \({\text{M}}_{\text{e}}\) across populations \(A\) and \(B\) based on 500 pre-selected SNPs (calculated from real genotype data) = 280; \({\text{M}}_{\text{e}}\) across populations \(A\) and \(B\) based on 48,412 remaining SNPs (calculated from real genotype data) = 32,970; genetic correlations between populations \(A\) and \(B\) = 0.6; heritability of the trait = 0.3 in both populations; proportion of genetic variance explained by SNPs in \({\mathbf{GRM}}_{1}\) = 0.5; number of individuals from population \(A\) in the training population = 476; number of individuals from population \(B\) in the training population = 5553