BAYESIAN SEMIPARAMETRIC ANALYSIS FOR TWO-PHASE STUDIES OF GENE-ENVIRONMENT INTERACTION

Area of Science:

• Genetics and Biostatistics
• Computational Biology
• Epidemiology

Background:

• Two-phase sampling designs are cost-efficient for collecting genetic data, especially for gene-environment (G x E) interaction analysis.
• Stratified sampling in phase II enriches for individuals with specific disease and exposure statuses, crucial for G x E studies.
• Existing methods for analyzing two-phase data may not fully leverage all available information.

Purpose of the Study:

• To develop a Bayesian analysis framework for two-phase studies investigating gene-environment interactions.
• To address challenges in high-dimensional models with multiple genes and environmental factors.
• To compare the proposed Bayesian method with existing weighted and pseudo-likelihood approaches.

Main Methods:

• Utilized a joint retrospective likelihood for phase I and phase II data.
• Employed a Bayesian variable selection algorithm for high-dimensional models.
• Incorporated hierarchical priors reflecting gene-gene and gene-environment independence assumptions.
• Modeled joint distribution of categorical variables using non-parametric Bayes.

Main Results:

• The proposed Bayesian method demonstrated advantages in characterizing joint effects of genotype and exposure.
• The approach efficiently utilized all available data from both phases.
• Simulation studies indicated superior performance compared to weighted and pseudo-likelihood methods.

Conclusions:

• The Bayesian approach offers a robust and efficient strategy for analyzing gene-environment interactions in two-phase studies.
• This method effectively handles complex models and high-dimensional genetic data.
• The findings are applicable to epidemiological studies, such as the colorectal cancer example involving statin use and genetic markers.