Sparse CCA-based mediation analysis with high-dimensional exposures and mediators

Area of Science:

• Environmental Health Sciences
• Biostatistics
• Computational Biology

Background:

• Mediation analysis is vital for understanding environmental exposures' effects on health via intermediate variables.
• High-dimensional data in environmental studies necessitates advanced methods for separating direct and indirect effects.
• Existing methods struggle with complex scenarios involving numerous exposures and mediators.

Purpose of the Study:

• To propose a novel mediation analysis method tailored for high-dimensional environmental exposure and mediator data.
• To accurately identify direct and indirect effects in complex environmental health studies.
• To evaluate the method's performance using simulations and real-world data.

Main Methods:

• Developed a mediation analysis framework utilizing Sparse Canonical Correlation Analysis (SCCA).
• Incorporated a two-step screening extension for enhanced feature selection and stable estimation.
• Applied the method to analyze exposure-metabolite pathways linked to MELD score.

Main Results:

• The SCCA-based method successfully identified relevant mediators and pathways in simulations, especially in high-dimensional, noisy settings.
• The two-step screening improved feature selection and maintained estimation stability.
• Real-data analysis revealed interpretable exposure-metabolite pathways associated with MELD score, demonstrating robustness.

Conclusions:

• The proposed SCCA-based mediation framework is effective for high-dimensional environmental health data.
• The method aids in identifying key exposure-mediator-outcome pathways.
• The approach offers a robust tool for environmental epidemiology and biomarker discovery.