A Computationally Efficient Algorithm for Genomic Prediction Using a Bayesian Model

Abstract Text:

Genomic prediction from dense SNP genotypes is widely used to predict breeding values for livestock, and crop breeding.  In many cases the most accurate methods are Bayesian, usually implemented via Markov Chain Monte Carlo (MCMC) scheme but this is computationally expensive. To retain the advantages of the Bayesian methods, with greatly reduced computation time, an efficient Expectation-Maximisation algorithm termed emBayesR is proposed. emBayesR retains the BayesR model’s prior assumption for SNP effects, of four normal distributions with increasing variance.  To improve the accuracy of genomic prediction compared to other non-MCMC approaches, emBayesR estimates the effect of each SNP while allowing for the error associated with estimation of all other SNP effects. Compared with BayesR, emBayesR reduced computational time up to 8 fold while maintaining similar prediction accuracy on both simulated data, and real 800K dairy data.

Keywords:

Genomic prediction

Expectation-Maximisation algorithm

Markov Chain Monte Carlo