Multivariate analyses using two genomic relationship matrices to weight predictive SNP markers

Abstract

<p>The Neogen GGP Ovine 50k chip contains approximately 5000 predictive Single-nucleotide polymorphisms (SNPs) that were identified by the Sheep CRC based on their relationship with carcase traits from genome wide association studies. These SNPs have been used in routine MERINOSELECT and LAMBPLAN analyses, equally-weighted with all other SNPs in a single genomic relationship matrix (GRM). This study aimed to examine the impact of fitting all SNPs in one GRM or fitting two GRMs, one with selected predictive SNPs and one with random SNPs, in conjunction with a numerator relationship matrix. Phenotypes on terminal sire breed cross resource flock animals recorded for five carcase and eating quality traits were used for bivariate variance component estimation. Variance components estimates were obtained for models containing only a numerator relationship matrix (NRM), NRM plus a GRM containing only non- selected SNPs, an NRM plus two GRMs containing non- selected and selected SNPs and an NRM plus one GRM containing all SNPs. Log-likelihoods were significantly higher in the models containing two GRMs for all trait pairs. Slightly higher average heritabilities were estimated from the model where the GRM contained all SNPs, except for intramuscular fat and shear force, where the GRM without the predictive SNPs resulted in higher heritabilities. The proportion of genetic variance explained by the genomic relationship matrices (𝜆) was estimated to be between 0.59 and 0.86. In terms of the genetic correlations between traits, for many trait-pairs the correlations were similar between the random effects fitted, but for two trait-pairs large differences were observed between the genetic correlation.</p>