Robust Bayesian variable selection for gene-environment interactions.

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Gene-environment (G×E) interactions are crucial for understanding complex diseases.
• Existing G×E studies often face challenges with outliers and data contamination in phenotypes.
• Robust regularization methods exist, but Bayesian approaches for this issue are underdeveloped.

Purpose of the Study:

• To develop a fully Bayesian robust variable selection method for G×E interaction studies.
• To address the challenge of outliers and heavy-tailed errors in disease phenotypes within a Bayesian framework.
• To enable robust identification of important main and interaction effects in G×E analyses.

Main Methods:

• A fully Bayesian robust variable selection method was developed.
• The method accommodates heavy-tailed errors and outliers in the response variable.
• Spike-and-slab priors were imposed at individual and group levels for robust sparse group selection.
• An efficient Gibbs sampler was implemented for fast computation.

Main Results:

• The proposed Bayesian method effectively handles outliers and data contamination.
• It demonstrated superior performance in identifying significant gene-environment interactions.
• The method showed robust variable selection capabilities in simulation studies and real-world data analyses.

Conclusions:

• The developed Bayesian method offers a robust approach for G×E interaction studies, particularly in the presence of data outliers.
• It provides a valuable tool for elucidating the etiology of complex diseases by accurately identifying genetic and environmental contributions.
• The method's effectiveness was validated through simulations and analyses of diabetes and melanoma datasets.