Estimation of total mediation effect for high-dimensional omics mediators.

Area of Science:

• Environmental health
• Genetics
• Biostatistics

Background:

• Environmental exposures influence health via molecular phenotypes like gene expression.
• Investigating high-dimensional intermediate phenotypes is crucial for understanding environmental health impacts.
• Mediation analysis is key, but lacks robust measures for total effects in high-dimensional settings.

Purpose of the Study:

• To extend the R-squared (R²) effect size measure for total mediation in moderate- and high-dimensional settings.
• To develop a reliable measure for the total mediation effect within a mixed model framework.
• To address the limitations of existing mediation measures in complex biological data.

Main Methods:

• Extended an R-squared (R²) effect size measure to high-dimensional mediator settings.
• Utilized mixed model framework for mediation analysis.
• Incorporated variable selection procedures (iterative sure independence screening, false discovery rate control) to identify relevant mediators.
• Developed a resampling-based confidence interval for estimation consistency.

Main Results:

• The proposed R²-based measure demonstrated small bias and variance in simulations.
• Variable selection procedures effectively excluded non-mediators, mitigating bias.
• The method was applied to the Framingham Heart Study, revealing gene expression profiles explain 38% of age-related systolic blood pressure variation.
• An R package "RsqMed" is available for practical application.

Conclusions:

• R-squared (R²) is an effective and efficient measure for total mediation effects, particularly in high-dimensional scenarios.
• The developed method provides a robust approach to analyze complex environmental exposures and molecular phenotypes.
• This research enhances our understanding of gene expression's role in mediating environmental impacts on health traits.