A maximum entropy model applied to spatial and temporal correlations from cortical networks in vitro

Area of Science:

• Computational neuroscience
• Systems neuroscience

Background:

• Multineuron firing patterns are often observed, but independent firing models predict them to be rare.
• Second-order maximum entropy models, using only firing rates and pairwise interactions, successfully predicted network correlations in retinal tissue.

Purpose of the Study:

• To assess the applicability of second-order maximum entropy models to cortical circuits.
• To investigate the temporal evolution of correlated network states in the cortex.

Main Methods:

• Applied a second-order maximum entropy model to multielectrode data (spikes and local field potentials) from cortical slices and cultures.
• Compared model predictions of network correlations with observed data.
• Analyzed the temporal sequences of correlated states and their relationship with pairwise temporal correlations.

Main Results:

• The model accounted for 88% of network correlations in cortical tissue, slightly less than in retinal tissue.
• Observed sequences of correlated states were significantly longer than predicted in 8 of 13 preparations, indicating temporal dependencies.
• A significant relationship was found between strong pairwise temporal correlations and observed sequence length.

Conclusions:

• Second-order maximum entropy models can predict correlated states in cortical networks but should be extended to incorporate temporal dynamics.
• Temporal dependencies are a common feature of cortical network activity and require consideration in future modeling efforts.
• Pairwise temporal correlations may provide a basis for extending these models into the temporal domain.