Fast parallelized sampling of Bayesian regression models for whole-genome prediction.

Area of Science:

• Genomics
• Computational Biology
• Statistical Genetics

Background:

• Bayesian regression models are crucial for genomic prediction, estimating marker effects simultaneously.
• Current methods rely on Markov chain Monte Carlo (MCMC), which is computationally intensive for whole-genome analyses.
• Tens of thousands of steps are often required, limiting practical application.

Purpose of the Study:

• To introduce a fast, parallelized algorithm for Bayesian regression models: independent intensive Bayesian regression models (BayesXII).
• To demonstrate how marker effect sampling can be made independent within MCMC steps for acceleration.
• To address the computational burden of Bayesian genomic prediction.

Main Methods:

• Developed the BayesXII algorithm, which augments the marker covariate matrix for orthogonal columns.
• Enabled independent sampling of marker effects within each MCMC step.
• Leveraged parallel computing, with potential acceleration up to 'p' times (number of markers).

Main Results:

• BayesXII was demonstrated using the BayesCπ prior on simulated data (50,000 individuals, 50,000 markers).
• Achieved similar accuracy to conventional samplers for BayesCπ in under 30 minutes using 24 nodes/cores.
• BayesXII was ten times faster than conventional samplers for BayesCπ.

Conclusions:

• The BayesXII algorithm significantly reduces computation time for Bayesian genomic prediction.
• Parallelization of Bayesian methods via BayesXII can increase their adoption and utility.
• This approach addresses the computational bottleneck in whole-genome analyses.