Meta-model for genomic relationships of metafounders applied on large scale single-step random regression test-day model

Abstract

In this study traditional genetic groups and metafounders were compared in analysis of a random regression TD model with ssGTBLUP.  The compared models were 1) ssGTBLUP with QP transformation of genetic groups and a genomic relationship matrix G built using base population allele frequencies for the markers, 2) ssGTBLUP with QP transformation of genetic groups and G with the same allele frequency of 0.5 for the markers, 3) ssGTBLUP with the metafounder (MF) approach and G with the same allele frequency 0.5.  All models used VanRaden method 1 in G and had a 30% residual polygenic proportion (RPG). The G matrix in cases 1) and 2) was scaled to have average diagonal equal to the pedigree-based relationship matrix A₂₂ of genotyped animals. Models 2) and 3) gave very similar results in terms of overprediction. Also, it seems that ssGTBLUP is quite robust to allele frequency used in the G matrix. However, the MF approach might be more efficient in reducing bias. In conclusion, both the QP transformation and the MF approach can be implemented in large-scale ssGTBLUP evaluation.