Variational Bayesian Method to Estimate Variance Components

Abstract Text:   The linear mixed model has been used in genome-wide association studies to account for population relatedness using variance component estimates. Here, we developed a Bayesian analysis using a variational inference method, a so-called variational Bayes (VB), to determine the marginal distributions of variance components. In addition to developing the VB method, we used simulation to compare its estimates of variance components with those from Gibbs sampling and restricted maximum likelihood. In the simulation scenarios we configured three levels of heritabilities (0.01, 0.50, and 0.80) and three population sizes (100, 500, and 5000). In the case of low heritability (0.01) and low population size (100), we observed strong bias in the VB method; less bias was detected with larger population sizes in every method examined. The computing time required was much shorter with the VB method than with Gibbs sampling.

Keywords:

variational Bayes

linear mixed model

variance component