Bayesian network mediation analysis with application to the brain functional connectome.

Area of Science:

• Neuroscience
• Biostatistics
• Network Science

Background:

• The brain's functional connectome is crucial for understanding brain function and mediating effects between exposures and outcomes.
• Current mediation analysis methods are limited, primarily focusing on univariate or vector mediators, neglecting network-level effects.

Purpose of the Study:

• To propose an integrative mediation analysis framework using a Bayesian paradigm to incorporate network-variate mediators.
• To address the methodological gap in analyzing mediation effects within complex brain network structures.

Main Methods:

• Introduced individually specified stochastic block models for parameterizing network measurements.
• Utilized latent network mediators derived from connectivity weights across network modules.
• Implemented simultaneous feature selection across network mediators to identify active components.

Main Results:

• Demonstrated the model's superiority in estimating various effect components.
• Showcased the model's effectiveness in selecting active mediating network structures.
• Successfully characterized the mediation of therapeutic intervention effects on opioid abstinence by brain functional sub-networks.

Conclusions:

• The proposed Bayesian integrative mediation analysis effectively handles network-variate mediators.
• This novel approach enhances understanding of complex mediation pathways in network neuroscience.
• The method provides a powerful tool for identifying critical brain network mediators in health and disease.