Adjusting the Genomic Relationship Matrix for Breed Differences in Single Step Genomic Blup Analyses

Abstract

The genomic relationship matrix (GRM) routinely constructed for single-step genomic BLUP analyses is known to contain breed structure, observable via principal component analysis, while the pedigree relationship matrix uses coefficients that are constant between known relatives regardless of breed or genetic group membership. This paper explores the effect of using allele frequencies for each breed or genetic group when calculating the GRM to reduce breed or genetic group structures in the GRM in the presence of pedigree based genetic groups fitted as random effects. We investigated the effect of using a breed-adjusted GRM on estimated breeding values, showing cross-validation results, genetic trends and estimated breeding value accuracies. Cross-validation results across breed showed a slight increase in EBV accuracy using a breed-adjusted GRM, 0.0220 ± 0.068 compared to a non-adjusted GRM, 0.206 ± 0.071. Genetic trends calculated from estimated breeding values (EBVs) using a breed-adjusted GRM were more closely aligned to those estimated using a pedigree-only model compared to a non-adjusted GRM. These results show that using a single set of allele frequencies in a GRM with a diverse number of breeds can result in biased breeding values and biased genetic trends relative to those obtained from pedigree model including breed groups.