Speeding up interval estimation for R 2 -based mediation effect of high-dimensional mediators via cross-fitting

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• Mediation analysis examines how molecular phenotypes, like gene expression, link exposures to health outcomes.
• Traditional mean-based measures can be inaccurate with high-dimensional omics data due to effect cancellation.
• Nonparametric bootstrap for R-squared measure is computationally intensive.

Purpose of the Study:

• To develop a computationally efficient estimation procedure for the variance-based R-squared mediation effect measure.
• To improve the accuracy and reduce the computational burden of mediation analysis in high-dimensional omics studies.
• To provide a reliable method for identifying gene expression mediation in complex health outcomes.

Main Methods:

• A two-stage, cross-fitted estimation procedure was developed for the R-squared measure.
• Iterative Sure Independence Screening (iSIS) was used in subsamples to identify true mediators and exclude non-mediators.
• Ordinary least squares regressions estimated variance, and confidence intervals were derived from the asymptotic distribution of the R-squared measure.

Main Results:

• The proposed procedure significantly enhances computational efficiency compared to resampling methods while maintaining comparable coverage probability.
• Simulations confirmed the method's accuracy and efficiency.
• Application to the Framingham Heart Study validated findings on gene expression mediating systolic blood pressure and identified gene roles in sex and HDL cholesterol relationships.

Conclusions:

• The novel two-stage, cross-fitted procedure offers a computationally efficient and accurate alternative for estimating R-squared mediation effects.
• This method is valuable for high-dimensional omics data, overcoming limitations of traditional approaches.
• The R package CFR2M implements this efficient procedure for broader research application.