Benchmarking Bayesian genome enabled-prediction models for age at first calving in Nellore cows

Abstract

Cow fertility traits are key factors that influence beef production profitability, and is particularly important in tropical environments where achieving high reproductive rates is challenging. Genomic selection (GS) has the potential to improve genetic gain rates for reproduction, if genomic estimated breeding values (GEBV) for these traits are sufficiently accurate. Several Bayesian models have already been proposed for GS, but the benchmarks used to compare them are still scarce, mainly for age at first calving (AFC) in Nellore cattle. A total of 714 AFC records of Nellore cows and 70 K SNPs were used to compare five models, Bayes A (BA), Bayes B (BB), Bayes Cπ (BCπ), Bayesian LASSO (BL) and Bayesian Ridge Regression (BRR). These models were compared by cross validation, randomly partitioning the whole population into 7 subsets (7-fold) and replicated 15 times. The prediction accuracy were 0.24 (0.11), 0.23 (0.11), 0.33 (0.13), 0.24 (0.11) and 0.38 (0.13), for BA, BB, BCπ, BL and BRR, respectively. Thus, BRR resulted in 14%, 15%, 5% and 14% additional prediction accuracy compared to BA, BB, BCπ and BL, respectively. Pearson and Spearman correlations between GEBVs obtained from BRR and BB models were, 0.97 and 0.94, respectively. It suggested that little difference in terms of animal selection would result from these methods. A more parsimonious model, such as BRR, can be successfully used in breeding programs to generate GEBVs which further enable consistent selection decisions. Although moderate accuracies of GEBV for AFC can be achieved, we found low efficiency of GS for AFC in the present population due to the small sample size and low heritability, reinforcing that GS efficiency is highly dependent upon these factors.