TESTING FOR THE CAUSAL MEDIATION EFFECTS OF MULTIPLE MEDIATORS USING THE KERNEL MACHINE DIFFERENCE METHOD IN...

Area of Science:

• Genomics
• Epigenetics
• Biostatistics

Background:

• High-throughput genomic and epigenomic data offer new avenues for understanding disease mechanisms beyond traditional association studies.
• Causal mediation analysis is a framework for exploring indirect effects in biological pathways.
• DNA methylation in neighboring probes often acts collaboratively, necessitating methods that jointly model their effects.

Purpose of the Study:

• To develop a robust statistical method, kernel machine difference (KMD), for detecting mediation effects of multiple DNA methylation probes.
• To extend single mediator analysis to jointly model the mediatory role of neighboring methylation probes.
• To investigate the mediatory role of DNA methylation in the causal pathway between smoking behavior and lung function using the Normative Aging Study (NAS) data.

Main Methods:

• Developed the kernel machine difference (KMD) method based on the causal mediation analysis framework.
• Employed kernel machine regression to flexibly model the effects of multiple mediators on the outcome.
• The KMD test does not require explicit modeling of mediator-exposure or mediator-mediator correlations, enhancing robustness.

Main Results:

• The KMD method provides a robust and computationally efficient testing procedure for joint natural indirect effects (NIE).
• Simulations demonstrate that KMD gains robustness and power, especially for nonlinear effects.
• Application to the Normative Aging Study (NAS) data investigates DNA methylation's role in smoking and lung function.

Conclusions:

• The KMD method is a powerful tool for analyzing high-dimensional genomic regions in mediation analysis.
• This approach advances the understanding of biological mechanisms underlying diseases by exploring complex epigenomic data.
• The findings highlight the utility of KMD in real-world epigenome-wide association studies.