Summary and Implications
Typical implementations of genomic prediction utilize Markov chain Monte Carlo (MCMC) sampling to estimate effects. Metropolis-Hastings (MH) is a commonly-used algorithm. We considered three different Gibbs samplers to speed up BayesB, a commonly-used model for genomic prediction. These differ in the manner they sample the marker effect, the locus-specific variance and the indicator variable. They are a single-site Gibbs Sampler, a blocking Gibbs Sampler and a Gibbs Sampler with pseudo prior. These three versions of BayesB are about twice as fast as the one using a MH algorithm.
Iowa State University
Cheng, Hao; Fernando, Rohan L.; and Garrick, Dorian J.
"Three Different Gibbs Samplers for BayesB Genomic Prediction,"
Animal Industry Report:
AS 660, ASL R2867.
Available at: https://lib.dr.iastate.edu/ans_air/vol660/iss1/32