Studies on inflation of GEBV in single-step GBLUP for type

Abstract

The purpose of this study was to evaluate sources of inflation of GEBV in single-step GBLUP (ssGBLUP) evaluations. Tests involved 10 102 702 records of 18 type traits from 6 930 618 Holstein cows. A total of 576k animals with genotypes were used in the analyses and included 23 174 sires, 27 215 cows and 49 611 young animals. The genomic relationship matrix (G) was scaled for compatibility with the pedigree matrix for genotyped animals (A₂₂). Genomic estimated breeding values (GEBV) using phenotypes up to 2010 or up to 2014 were calculated considering a) inbreeding in A₂₂ but not in A, 2) inbreeding in both A and A₂₂, 3) as previously but considering nonzero inbreeding of phantom parents, and 4) as previously but reducing the additive variance by one half. Reliabilities (R²) and regression factors (b₁) were derived based on DYD2014 and GEBV2010 of 1711 sires with at least 50 daughters in 2014 but no daughters in 2010. For cases 1 to 4, the reliabilities were 0.48, 0.49, 0.49, and 0.50, respectively. The average regressions were 0.75, 0.85, 0.90, and 0.96, respectively. The b₁ factors could be close to 1 by multiplying G^-1 - A₂₂^-1 by lambda and by multiplying A₂₂^-1 by omega. The optimal omega was 0.7 for case 1) and 0.9 for case 2). Both parameters are shown to account for ignored inbreeding in A. Inflation of GEBV in ssGBLUP can be eliminated by considering inbreeding in A, correction factors, and by reducing the additive variance. A comprehensive theory extending unknown parent groups to metafounders may automatically eliminate inflation of GEBV for arbitrarily complex populations including multibreed.