Bayesian adaptive Markov chain Monte Carlo estimation of genetic parameters

Area of Science:

• Quantitative genetics
• Statistical genetics
• Computational biology

Background:

• Accurate estimation of genetic parameters is crucial for quantitative trait analysis in natural and breeding populations.
• Linear mixed models incorporating additive and dominance effects are commonly used.
• Existing methods may lack speed or efficiency in parameter estimation.

Purpose of the Study:

• To develop a novel, fast adaptive Markov chain Monte Carlo (MCMC) sampling algorithm.
• To improve the estimation of genetic parameters within linear mixed models.
• To enhance the analysis of quantitative traits.

Main Methods:

• Proposed a hybrid Gibbs sampler for learning the covariance structure of variance components.
• Developed a Metropolis-Hastings algorithm utilizing the learned covariance structure for an effective proposal distribution.
• Integrated out random effects in the likelihood function.

Main Results:

• The new adaptive MCMC algorithm demonstrated superior mixing properties compared to the hybrid Gibbs sampler.
• The algorithm ran approximately twice as fast as the hybrid Gibbs sampler.
• Posterior mode estimates closely aligned with residual maximum likelihood (REML) estimates.
• Successfully detected multiple modes in the posterior distribution.
• Vague exponential prior allowed for posterior variance estimates near zero when supported by data.

Conclusions:

• The developed adaptive MCMC algorithm offers a faster and more efficient approach for estimating genetic parameters in mixed linear models.
• This method provides accurate estimates and robust detection of posterior distribution modes.
• The algorithm shows promise for applications in quantitative genetics, breeding, and population genetics research.