Prediction of Genomic Breeding Values for Binary Traits Using Variational Bayes Methods

Numerous Bayesian methods for estimating SNP effects in the prediction of genomic breeding value have been developed, and these methods have used a Markov chain Monte Carlo (MCMC) algorithm. However, the MCMC algorithm for binary or polychotomous response traits using data argumentation has convergence problems, such as slow mixing. In this study, we proposed the variational Bayes algorithms for the three genomic evaluation methods, the so-called BayesA, BayesB and Bayesian ridge-regression, combined with threshold model for binary traits. Variaional Bayes is a kind of deterministic method for constructing approximate posterior distributions for parameters. We evaluated the accuracies of the proposed variatonal Bayes algorithms for estimating genomic breeding values using simulation datasets. The results indicated that the variational Bayes methods could be an actual choice for estimating the genomic breeding values of binary traits.