Bayesian Joint Modeling of Multiple Brain Functional Networks.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Statistical Modeling

Background:

• Analyzing brain network changes across different mental conditions is crucial in neuroscience.
• Standard methods often lack the accuracy and power to detect subtle differences between networks.
• Existing approaches struggle to pool information effectively across multiple related networks.

Purpose of the Study:

• To develop an integrative Bayesian approach for jointly modeling multiple brain networks.
• To provide a robust inferential framework for comparing brain networks under varying mental conditions.
• To enhance the detection of shared and differential patterns in brain connectivity.

Main Methods:

• Developed a novel Bayesian approach using Dirichlet process priors for edge probabilities.
• Employed Bayesian spike and slab priors on the precision matrix off-diagonals for connection strengths.
• Validated the method through numerical simulations and analysis of fMRI Stroop task data.

Main Results:

• The proposed Bayesian method demonstrated increased power in detecting differential edges with controlled false positives.
• Achieved superior brain network estimation accuracy compared to existing separate estimation strategies.
• fMRI Stroop task analysis revealed significant connectivity differences between task and fixation states, and nuanced differences between exertion and relaxation.

Conclusions:

• The integrative Bayesian approach offers a powerful and accurate framework for brain network comparison.
• It effectively identifies condition-specific connectivity changes, providing more biologically plausible results than penalized methods.
• This method advances the understanding of dynamic brain network alterations during cognitive tasks.