Evaluating the impact of including residual polygenic effects in dairy genomic evaluations using Bayesian methods

Abstract

The impact of different levels of residual polygenic effect was studied for milk yield and somatic cell count (SCC) using GBLUP, BayesA and BayesB in a Holstein/Friesian population. The reference population consisted of 8605 and 7092 bulls with at least 10 EDCs and reliability of 69% for milk yield and SCC respectively. Corresponding bulls in the validation set were 4090 and 2448 respectively. A linear model was used for the estimation of SNP effects with fixed mean effect and random residual polygenic (RP) and SNP effects. The RP levels were set at 0, 5, 10, 15, 20 and 25% of the total genetic variance.  In the case of BayesA and BayesB, analyses were carried out with no restriction on the percentage of the total genetic variance accounted for by the polygenic effect. The regressions of direct genomic breeding values (DGV) on de-regressed sire proofs (DSP) increased (0.47 to 0.98) with increasing levels of polygenic effect for both traits.  The regressions for milk yield were similar for GBLUP and BayesB but higher than estimates from BayesA, indicating slightly poorer predictions from BayesA. Similarly, higher regressions were obtained for SCC from GBLUP compared to BayesA. In general correlations between DGV and DSP were generally higher with GBLUP compared with the Bayesian methods at each level of RP fitted. The correlations of the polygenic solutions with DGV were always higher for SCC at every level of RP compared with milk yield indicating the higher impact of the RP effect for traits of lower heritability.  Contributions from parent average accounted for 68% and 89% of the RP contributions for bulls with at most 15 or 100 EDCs with GBLUP at 10% RP for milk yield and SCC respectively. There was a decreasing trend in the mean of SNP solutions for SNPs with alleles of medium, and high frequencies as the percentage of polygenic increased with all three methods for milk yield. However such a trend was not observed for SCC. In general mean SNP variances for milk yield and SCC declined with increasing levels of polygenic effects. When no constraint was imposed on the level of RP, estimates of the polygenic variance were unexpectedly high from BayesA and BayesB which were difficult to interpret.