A guide to group effective connectivity analysis, part 2: Second level analysis with PEB.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Neuroscience

Background:

• Characterizing inter-subject variability in neural circuitry is crucial for understanding brain function.
• Existing methods may not fully capture the nuances of effective connectivity across diverse populations.
• Dynamic Causal Modelling (DCM) and Parametric Empirical Bayes (PEB) offer advanced frameworks for this analysis.

Purpose of the Study:

• To provide a detailed tutorial and worked example of using DCM and PEB for characterizing inter-subject variability in effective connectivity.
• To explain the underlying theory, assumptions (priors), and hierarchical modeling approach.
• To demonstrate group-level analysis procedures applicable to various neuroimaging modalities.

Main Methods:

• A hierarchical modeling approach with multiple levels was employed.
• State-space models (DCMs) were used at the first level to infer subject-specific effective connectivity from neuroimaging time-series (fMRI, MEG, EEG).
• A General Linear Model (GLM) was applied at the group level to model subject-specific parameters, partitioning variability and enabling hypothesis testing.

Main Results:

• The Bayesian hierarchical model effectively conveys estimated connection strengths and their uncertainty from the subject to the group level.
• The approach allows for testing hypotheses about commonalities and differences in effective connectivity across subjects.
• Group-level parameters can refine subject-level parameter estimation by serving as empirical priors.

Conclusions:

• The combined DCM and PEB approach provides a powerful framework for analyzing inter-subject variability in effective connectivity.
• This methodology is adaptable to various neuroimaging data types and facilitates robust inference on neural circuitry.
• The provided worked example and dataset enable reproducible research in computational neuroscience.