Estimation of quantitative trait locus effects with epistasis by variational Bayes algorithms

Area of Science:

• Genetics
• Statistical Genomics
• Computational Biology

Background:

• Bayesian hierarchical shrinkage methods are standard for quantitative trait locus (QTL) mapping.
• Markov chain Monte Carlo (MCMC) is computationally intensive for high-dimensional genetic analyses, like epistatic interactions.
• Maximum a posteriori (MAP) estimation offers speed but lacks uncertainty quantification.

Purpose of the Study:

• To propose variational Bayes (VB) as an efficient computational approach for Bayesian hierarchical shrinkage models in QTL mapping.
• To evaluate VB's ability to provide uncertainty estimates, bridging the gap between MAP and MCMC.
• To compare the performance of VB-based methods against MCMC counterparts and existing R packages.

Main Methods:

• Implementation of variational Bayes algorithms for three hierarchical shrinkage models: Bayesian adaptive shrinkage, Bayesian LASSO, and extended Bayesian LASSO.
• Utilizing posterior credible intervals and permutation tests for QTL detection decisions.
• Comparative analysis using a simulated public epistatic dataset against R/qtlbim and R/BhGLM.

Main Results:

• Variational Bayes methods demonstrated strong performance in quantitative trait locus mapping.
• VB approaches were found to be highly competitive with traditional MCMC methods.
• The proposed VB models offer a computationally efficient alternative with uncertainty estimation.

Conclusions:

• Variational Bayes provides an effective and efficient computational strategy for Bayesian QTL mapping, especially for complex genetic architectures.
• VB methods successfully integrate uncertainty estimation, enhancing decision-making in genetic analyses.
• The proposed VB framework offers a valuable extension to existing Bayesian statistical genetics tools.