Joint high-dimensional Bayesian variable and covariance selection with an application to eQTL analysis.

Area of Science:

• Genetics
• Statistical Modeling
• Bioinformatics

Background:

• High-dimensional data in genetics presents challenges for variable selection and association analysis.
• Existing methods struggle with the complexity of correlated responses and numerous predictors.
• Understanding gene expression regulation requires robust statistical frameworks.

Purpose of the Study:

• To develop a Bayesian technique for sparse joint selection of predictors and response variable interactions.
• To enable association analysis between high-dimensional predictors and responses in a sparse seemingly unrelated regression (SSUR) framework.
• To infer the interaction network of genetic transcripts while accounting for predictor effects.

Main Methods:

• A marginalization-based collapsed Gibbs sampler is employed for efficient model space exploration.
• Spike and slab priors are utilized to handle high-dimensional settings where predictors and responses exceed sample size.
• Bayesian inference is combined with Gaussian graphical models for conditional independence statements.

Main Results:

• The method successfully performs sparse joint selection of significant predictor variables and inverse covariance matrix elements.
• It enables efficient association analysis between high-dimensional sets of predictors and responses.
• The approach was applied to expression quantitative trait loci (eQTL) analysis, revealing significant SNP-transcript associations and transcript interaction networks.

Conclusions:

• The developed Bayesian technique offers a computationally feasible and efficient solution for high-dimensional SSUR.
• It effectively identifies genetic associations and facilitates the inference of biological networks.
• The method demonstrates superior performance compared to existing Bayesian approaches for similar analyses.