Advanced Bayesian kernel machine regression for large-scale exposome studies: Making the impossible possible.

Area of Science:

• Environmental Health Sciences
• Statistical Modeling
• Computational Biology

Background:

• Exposome studies face challenges with complex exposure interactions and collinearity using traditional methods.
• Bayesian kernel machine regression (BKMR) is promising but limited by computational cost and interpretability.

Purpose of the Study:

• To develop an advanced BKMR (A-BKMR) model for efficient and interpretable analysis of large-scale exposome data.
• To overcome the computational and interpretability limitations of conventional BKMR.

Main Methods:

• Implemented Gaussian predictive process and matrix decomposition for reduced processing time and memory.
• Utilized the parametric g-formula for interpretable joint, univariate, and interaction effect statistics.

Main Results:

• A-BKMR demonstrated high computational efficiency, analyzing datasets of 100,000 samples in 1 hour (over 700,000x faster than traditional BKMR).
• Achieved high model performance with Area Under the Curve (AUC) > 0.99 and R-squared > 0.97.
• Introduced novel quantitative metrics for enhanced interpretability of effect and interaction analyses.

Conclusions:

• A-BKMR provides a computationally efficient and highly interpretable statistical framework for large-scale exposome studies.
• The model accurately identifies key exposures and their interactions, facilitating deeper understanding of environmental health impacts.