A novel strategy for detecting multiple mediators in high-dimensional mediation models.

Area of Science:

• Biostatistics
• Genomics
• Computational Biology

Background:

• High-dimensional mediation models often suffer from overestimated direct effects, leading to inaccurate identification of true mediators.
• Existing methods struggle with identifying multiple biomarkers simultaneously in complex biological systems.

Purpose of the Study:

• To develop a novel methodology for accurate detection of multiple biomarkers in high-dimensional mediation models.
• To address the overestimation of direct effects and improve the identification of mediators with significant indirect effects.

Main Methods:

• Utilized a modified Least Absolute Shrinkage and Selection Operator (LASSO) combined with Pathway LASSO.
• Introduced two constraints on the L1-norm penalty to mitigate direct effect overestimation.
• Incorporated sure independence screening for optimal dimension reduction threshold selection.

Main Results:

• The proposed methodology demonstrated superior performance in simulations across various scenarios compared to existing methods.
• Achieved improved true positive rates for mediator detection and enhanced accuracy in identifying true biomarkers.
• The method proved robust and reliable, suitable for real-world applications.

Conclusions:

• This novel approach offers a substantial advancement in biomarker detection for high-dimensional mediation analysis.
• The methodology enhances the accuracy and robustness of mediator identification, with significant implications for clinical research and practice.
• Demonstrated practical utility through application to datasets on internalizing psychopathology and late-life depression.