BHCox: Bayesian heredity-constrained Cox proportional hazards models for detecting gene-environment interactions

Area of Science:

• Genetics and Epidemiology
• Biostatistics
• Computational Biology

Background:

• Gene-environment (G×E) interactions are crucial for disease etiology and prognosis.
• Detecting G×E interactions in censored survival data presents challenges like high dimensionality and complex environmental effects.
• Effect heredity principles have not been applied to Bayesian Cox models for censored survival outcome interactions.

Purpose of the Study:

• To propose novel Bayesian heredity-constrained Cox proportional hazards (BHCox) models.
• To incorporate effect heredity into Bayesian Cox models for detecting G×E interactions in censored survival data.
• To identify and estimate main and interaction effects using advanced Bayesian priors.

Main Methods:

• Developed BHCox models utilizing spike-and-slab and regularized horseshoe priors.
• Employed the no-U-turn sampler (NUTS) algorithm via the R package brms for model fitting.
• Conducted extensive simulations to evaluate BHCox model performance against alternative methods.

Main Results:

• BHCox models demonstrated superior performance compared to alternative models in simulation studies.
• The proposed method successfully identified biologically plausible gene-smoking interactions in non-small-cell lung cancer (NSCLC) patient data.
• Identified significant G×E interactions impacting NSCLC patient prognosis.

Conclusions:

• BHCox models effectively detect main effects and interactions in high-dimensional censored survival data.
• The developed method has significant implications for discovering complex G×E interactions.
• Provides a robust framework for analyzing G×E interactions in survival analysis.