Neural assemblies uncovered by generative modeling explain whole-brain activity statistics and reflect structural...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Brain activity arises from complex interactions within neural assemblies.
• Understanding how neuronal assembly dynamics generate large-scale brain activity patterns is crucial.
• Existing models often struggle to capture the full statistical properties of whole-brain recordings.

Purpose of the Study:

• To develop and validate a data-driven generative model capable of reproducing whole-brain activity statistics.
• To identify and characterize neural assemblies and their role in brain state transitions.
• To investigate interregional functional connectivity using in silico perturbations.

Main Methods:

• Simultaneous recording of activity from approximately 40,000 neurons in zebrafish larvae.
• Application of a compositional Restricted Boltzmann Machine (cRBM) as a generative model.
• In silico perturbation experiments to infer functional connectivity.

Main Results:

• The cRBM accurately reproduced mean activity and pairwise correlation statistics of spontaneous neural activity.
• Approximately 200 neural assemblies were identified, forming circuits whose combinations define brain states.
• In silico derived functional connectivity was conserved across individuals and correlated with structural connectivity.

Conclusions:

• The compositional Restricted Boltzmann Machine effectively captures the coarse-grained organization of the zebrafish brain.
• This generative modeling approach provides insights into neural assembly interactions and brain state dynamics.
• The cRBM framework is adaptable for analyzing large-scale neural data from various recording techniques.