A Hierarchical Bayesian Model for Differential Connectivity in Multi-trial Brain Signals

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Understanding brain connectivity is crucial for neuroscience research.
• Existing methods face challenges in modeling within- and between-condition connectivity variations.
• Accurate inference of effective connectivity across trials and conditions is needed.

Purpose of the Study:

• To propose a novel Bayesian hierarchical vector autoregressive (BH-VAR) model for characterizing brain connectivity.
• To infer differences in connectivity across experimental conditions and trials.
• To incorporate within-condition similarity and between-conditions heterogeneity in connectivity modeling.

Main Methods:

• Developed a Bayesian hierarchical vector autoregressive (BH-VAR) model.
• Utilized partial directed coherence (PDC) for frequency-specific effective connectivity inference.
• Applied a two-stage computation approach for efficient parameter estimation and uncertainty quantification.
• Analyzed local field potentials (LFPs) from rat hippocampus during a memory task.

Main Results:

• The BH-VAR model successfully characterized hippocampal connectivity during a memory task.
• Identified two distinct functional units within the CA1 region: lateral and medial segments.
• Observed stronger self-connectivity within each functional unit.
• Revealed a primary lateral-to-medial information flow direction within trials.
• Demonstrated condition-specific differences in this lateral-to-medial information flow.

Conclusions:

• The proposed BH-VAR model effectively quantifies variations in functional connectivity within and between conditions.
• The model provides novel insights into hippocampal network dynamics during memory encoding and retrieval.
• This approach offers broad applicability for analyzing complex brain connectivity data in neuroscience research.