Approximation of genomic accuracies in single-step genomic evaluation

Abstract

Reliability of predictions from single-step genomic BLUP (ssGBLUP) can be calculated by inversion, but that is not feasible for large data sets. Two proposed approximations of reliability are based on decomposition of a function of reliability into contributions from records, pedigree, and genotypes. The first approximation involves inversion of a matrix that contains inverses of the genomic relationship matrix (G) and the pedigree relationship matrix for genotyped animals (A22). The second approximation involves only the diagonal elements of those inverses. The approximations were tested with a simulated data set. The correlations between exact and approximated contributions due to genomic information were 0.92 for the first approximation and 0.56 for the second approximation; contributions were inflated 60 and 260%, respectively. The respective correlations for reliabilities were 0.98 and 0.72. After correction for inflation, those correlations increased to 0.99 and 0.89. Approximations of reliabilities of predictions by ssGBLUP are accurate and computationally feasible. A critical part of the approximations is quality control of SNP information and proper scaling of G.