High-dimensional Many-to-many-to-many Mediation Analysis

Area of Science:

• Biostatistics
• Genetics
• Neuroscience

Background:

• High-dimensional mediation analysis is crucial for understanding complex biological systems.
• Existing methods struggle with multivariate exposures, mediators, and outcomes simultaneously.
• The need for a framework handling many-to-many-to-many relationships is evident.

Purpose of the Study:

• To develop and validate a Many-to-Many-to-Many (MMM) mediation analysis framework.
• To enable simultaneous variable selection, indirect effect estimation, and outcome prediction in high-dimensional settings.
• To apply the MMM framework to genetic and neuroimaging data in Alzheimer's disease research.

Main Methods:

• Formalized the problem as Many-to-Many-to-Many (MMM) mediation analysis.
• Developed methods for simultaneous variable selection and indirect effect matrix estimation.
• Validated the framework through simulations and application to Alzheimer's Disease Neuroimaging Initiative data.

Main Results:

• The MMM mediation analysis framework demonstrates consistency and asymptotic normality.
• Simulation studies confirmed its finite-sample performance, convergence, and robustness.
• Identified significant many-to-many-to-many genetic-neural-cognitive pathways in Alzheimer's disease.
• Improved out-of-sample classification and prediction performance for cognitive and diagnostic outcomes.

Conclusions:

• The MMM mediation analysis provides a powerful tool for investigating complex, high-dimensional multi-layer pathways.
• The framework offers biologically interpretable insights into gene-brain-cognition relationships.
• Statistical methodology advancements are vital for complex scientific investigations.