Powerful large scale inference in high dimensional mediation analysis.

Area of Science:

• Genomics
• Epigenetics
• Biostatistics

Background:

• Identifying causal pathways from exposures to outcomes via intermediate variables (e.g., DNA methylation) is crucial in genome-wide epigenetic studies.
• Traditional mediation analysis faces challenges with complex null hypotheses and is often underpowered for high-dimensional data.
• Existing methods like Sobel's test and Max-P test are limited by suboptimal null distributions and failure to address multiple testing burdens.

Purpose of the Study:

• To develop a novel, powerful method for high-dimensional mediation analysis.
• To address the limitations of existing methods in identifying causal mediation effects in complex biological data.
• To improve the discovery of biologically relevant mediators in genome-wide studies.

Main Methods:

• Introduction of MLFDR (Mediation Analysis using Local False Discovery Rates), a new statistical framework.
• Utilizing local false discovery rates derived from structural equation model coefficients.
• Constructing an optimal rejection region for robust mediation effect testing.

Main Results:

• MLFDR theoretically and computationally demonstrates asymptotic control of the false discovery rate.
• The method achieves superior statistical power compared to existing high-dimensional mediation techniques.
• Real-world data applications showed MLFDR identified 20%-50% more significant mediators than conventional methods.

Conclusions:

• MLFDR offers a statistically rigorous and powerful approach for high-dimensional mediation analysis.
• The method enhances the ability to detect subtle biological signals missed by current techniques.
• MLFDR represents a significant advancement for causal pathway discovery in epigenome-wide association studies.