Robust sparse Bayesian regression for longitudinal gene-environment interactions

Area of Science:

• Biostatistics
• Genomics
• Statistical Genetics

Background:

• Longitudinal studies require accurate estimation of main and interaction effects.
• High-dimensional genetic data presents challenges for traditional analysis of variance (ANOVA).
• Sparse longitudinal gene-environment (G×E) interactions are understudied, especially with skewed data and correlated observations.

Purpose of the Study:

• To develop a novel robust sparse Bayesian mixed model for longitudinal gene-environment interaction analysis.
• To address challenges including skewed phenotypic measurements, intra-cluster correlations, and structured sparsity.
• To enable robust Bayesian variable selection for main and interaction effects.

Main Methods:

• Developed a robust sparse Bayesian mixed model incorporating structured spike-and-slab priors.
• Implemented Gibbs samplers and Markov Chain Monte Carlo (MCMC) algorithms for efficient computation and posterior inference.
• Accommodated outliers and inter-relatedness among repeated measurements.

Main Results:

• The proposed model demonstrated superior performance in variable selection and estimation compared to benchmark methods in extensive simulations.
• Successfully analyzed longitudinal lipidomics data from a cancer prevention study in CD-1 mice.
• Identified significant main and interaction effects with important biological implications.

Conclusions:

• The novel robust sparse Bayesian mixed model effectively handles challenges in high-dimensional longitudinal gene-environment interaction analysis.
• The method offers improved prediction performance over existing alternatives.
• Provides a powerful tool for uncovering complex genetic and environmental influences in biological studies.