Balanced neural architecture and the idling brain.

Area of Science:

• Computational neuroscience
• Systems neuroscience

Background:

• Cortical spike trains exhibit significant trial-to-trial variability, which differs between spontaneous and stimulus-driven states.
• Existing models of balanced recurrent cortical networks capture evoked variability but fail to replicate spontaneous dynamics.

Purpose of the Study:

• To investigate how global network architectures influence neural variability in recurrent cortical networks.
• To determine if attractor network models can simultaneously explain spontaneous and evoked cortical dynamics.

Main Methods:

• Simulated balanced spiking networks with different global architectures (clustered assembly, feedforward chain, ring structures).
• Analyzed trial-to-trial variability, signal correlations, and stimulus-induced changes in network dynamics.
• Contrasted attractor networks with strongly coupled chaotic networks.

Main Results:

• Attractor networks with global structures successfully replicate both spontaneous and evoked neural variability.
• Signal correlations in spontaneous states are linked to the network's global structure and stimulus preference.
• Only attractor networks exhibit stimulus-induced quenching of trial-to-trial variability, matching experimental observations.

Conclusions:

• Global network architecture, particularly attractor properties, is crucial for explaining diverse neural variability patterns in the cortex.
• Comparing neural dynamics across spontaneous and evoked states provides insights into the underlying global structure of cortical networks.