Mediation Analysis for High-Dimensional Mediators and Outcomes with an Application to Multimodal Imaging Data

Area of Science:

• Neuroimaging
• Brain Research
• Systems Neuroscience

Background:

• Multimodal neuroimaging data integration is crucial for understanding brain function and phenotypes.
• Analyzing complex interactions within multimodal data presents significant challenges.
• Existing methods struggle with the intricate relationships between multivariate imaging variables.

Purpose of the Study:

• To introduce a novel method, Multimodal Mediation Analysis (MMA), for integrated analysis of multimodal neuroimaging data.
• To simultaneously extract latent mediation patterns and estimate mediation effects.
• To address the challenges posed by complex interactive relationships in multimodal data.

Main Methods:

• Development of a novel Multimodal Mediation Analysis (MMA) approach.
• Utilizing a dense bi-cluster graph for pattern extraction and effect estimation.
• Implementing a computationally efficient algorithm for structure estimation and inference with multiple testing correction.

Main Results:

• MMA demonstrates superior performance compared to existing methods in simulations.
• MMA shows improved false discovery rate and sensitivity.
• The method was successfully applied to Human Connectome Project data.

Conclusions:

• MMA offers a robust framework for analyzing multimodal neuroimaging data.
• The method effectively identifies mediation patterns and effects in complex brain data.
• MMA advances the investigation of neural mechanisms underlying different phenotypes, exemplified by blood pressure's effect on brain imaging measures.