Bayesian mixture modeling of gene-environment and gene-gene interactions

Area of Science:

• Genetics
• Biostatistics
• Epidemiology

Background:

• Advancements in genotyping technology enable large-scale investigation of gene-environment and gene-gene interactions.
• High dimensionality in parameter space poses computational challenges and can lead to unstable models with wide confidence intervals.

Purpose of the Study:

• To develop a statistical framework for simultaneously investigating numerous gene-environment and gene-gene interactions.
• To enhance the power for detecting true genetic signals by effectively handling a large number of potential interactions.

Main Methods:

• A hierarchical mixture model is proposed to analyze interactions.
• The model utilizes a mixture prior with components for null and significant effects, effectively shrinking null effects towards zero.
• The framework is flexible, allowing incorporation of prior substantive information.

Main Results:

• The hierarchical mixture model allows simultaneous investigation of all interactions.
• The method increases statistical power for detecting real genetic signals by down-weighting small, null effects.
• Simulations and a lung cancer case-control study demonstrate the method's utility.

Conclusions:

• The developed hierarchical mixture model offers a powerful approach for analyzing complex genetic interactions in large-scale studies.
• This method addresses the computational challenges and statistical instability associated with high-dimensional interaction analysis.
• The flexible prior framework facilitates the integration of existing knowledge into genetic association studies.