Estimation of genomic breeding values in Gir cattle using Bayesian methods

Abstract

The paper evaluates different Bayesian methods in estimating genomic breeding values using a female reference population of Gir breed and assesses their predicting ability and compares the results achieved with those using Genomic Best Linear Unbiased Prediction (GBLUP). In the current study, BayesA, BayesB, BayesCπ, Bayes Lasso, and Bayes-GBLUP methods were used on the same dataset that employed GBLUP and compared the predicting ability of methods using the same five cross-validation data sets. Phenotypic standard 305-day lactation records of 2571 Gir cows extracted from the INAPH database were used in the current study. The milk yield records were pre-corrected for environmental effects. All 2571 cows were genotyped using INDUSCHIP developed by National Dairy Development Board (NDDB). The correlation of SNP effects between BayesCᴨ and Bayesian Lasso was 0.99 and their correlations to BayesA and BayesB were 0.814 and 0.685 respectively. The correlation of SNP effects of BayesA and BayesB was 0.977. The correlations between the predicted genomic breeding values and the corrected phenotypic values in all methods were about 0.89 and those between estimated genomic breeding values of different Bayesian methods were equal to one. On comparison of the correlations between estimated genomic breeding values and corrected phenotypic values of five cross-validation data sets for each of five Bayesian methods, it was observed that the Bayes-GBLUP performed better than GBLUP and also better than all other four Bayesian methods. The other four Bayesian methods BayesA, BayesB, BayesCᴨ, and BayesBL performed slightly better than GBLUP. The correlations among them were more or less the same. From the current study, it could be concluded that Bayesians methods are very useful in estimating both genetic marker effects associated with QTLs and genomic breeding values and that they could be applied in practice for selecting young calves without phenotypic records, based on their genomic breeding values with reasonably high confidence.