Model-free High Dimensional Mediator Selection with False Discovery Rate Control

2025-05-14

AI Summary

Controlling the false discovery rate (FDR) in high-dimensional mediation analysis is challenging when mediators exhibit complex dependencies and interactions. Method: This paper proposes a model-agnostic framework for composite null hypothesis testing, the first to formulate high-dimensional mediator selection as a sequential composite null testing problem—bypassing restrictive linear or sparsity assumptions and achieving theoretically guaranteed FDR control under weak modeling conditions. The method integrates multiple testing correction, dependency-robust p-value combination, bootstrap resampling, and an adaptive thresholding algorithm, supporting nonparametric effect estimation. Results: Simulations demonstrate substantial gains in statistical power and FDR stability. Applied to the ADNI dataset, the method successfully identifies key MRI mediators—including hippocampal and amygdalar volumes—and uncovers neuroimaging pathways through which sex modulates dementia progression.

Technology Category

• Machine Learning: Dimensionality Reduction/Feature Selection
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

There is a challenge in selecting high-dimensional mediators when the mediators have complex correlation structures and interactions. In this work, we frame the high-dimensional mediator selection problem into a series of hypothesis tests with composite nulls, and develop a method to control the false discovery rate (FDR) which has mild assumptions on the mediation model. We show the theoretical guarantee that the proposed method and algorithm achieve FDR control. We present extensive simulation results to demonstrate the power and finite sample performance compared with existing methods. Lastly, we demonstrate the method for analyzing the Alzheimer's Disease Neuroimaging Initiative (ADNI) data, in which the proposed method selects the volume of the hippocampus and amygdala, as well as some other important MRI-derived measures as mediators for the relationship between gender and dementia progression.

Problem

• Selecting high-dimensional mediators with complex correlation structures
• Controlling false discovery rate in mediator hypothesis testing
• Validating method on Alzheimer's Disease Neuroimaging Initiative data

Innovation

• Frames mediator selection as hypothesis testing
• Controls false discovery rate with mild assumptions
• Validated on Alzheimer's Disease Neuroimaging data