Improving single-step genomic prediction reliabilities for clinical mastitis in Nordic Red dairy cattle and Jersey by applying marker-specific weights

Abstract

The standard single-step genomic prediction assumes that all single nucleotide polymorphism (SNP) markers explain an equal amount of genetic variance. The true state may deviate from this assumption, and it has been suggested to consider SNP marker-specific weights when predicting genomic enhanced breeding values (GEBV). We hypothesized that the benefit may be more pronounced in low heritable traits and investigated this hypothesis using the udder health evaluations for Nordic Red (RDC) and Jersey (JER) dairy cattle. In the first step, we develop a standard single-step genomic prediction (ssGBLUP) model based on the currently used multiple-trait evaluation models, and estimated GEBVs. The models included four clinical mastitis (CM) traits, and five correlated traits, namely test-day somatic cell score (SCS) in 1^st, 2^nd, and 3^rd lactations, fore udder attachment and udder depth, and describes all additive genetic effects of an animal by one covariance function. Then, we investigated three alternative approaches, where we applied SNP-marker specific weights. The three approaches for SNP-marker weighting were: 1) a nonlinear method similar to BayesA, 2) the classical formula (2pqû²), and 3) the mean of SNP weights for every 20 adjacent SNP markers calculated based on 2pqû². To solve the models with SNP marker-specific weights, we applied the single-step SNPBLUP solver implemented in MiX99. We validated the models by forward validation where the last four years of the data were removed. The datasets for RDC and JER included 6.9 and 1.2 million animals of which 5.6 and 0.9 million cows had records, respectively. The number of genotyped animals was 125,789 and 64,777 for RDC and JER, respectively. We found a significant increase in prediction reliability for CM when applying SNP-marker specific weights. For instance, applying the 2pqû² weights compared to the standard ssGBLUP for SCS, the prediction reliability increased from 0.58 to 0.64 and from 0.61 to 0.56 for RDC and JER bulls, respectively. We found similar improvements in the prediction reliability for cows. In general, all weighing approaches improved prediction reliability, but the highest improvement was achieved by weighing the SNP-markers by 2pqû².