Maximum entropy models reveal the correlation structure in cortical neural activity during wakefulness and sleep

Maximum Entropy models can be inferred from large data-sets to uncover how local interactions can generate collective dynamics. Here, we employ such methods to investigate the characteristics of neurons recorded by multielectrode arrays in the cortex of human and monkey throughout states of wakefulness and sleep. Taking advantage of the separation of cells into excitatory and inhibitory types, we construct a model including this biological feature. By comparing the performances of Maximum Entropy models at predicting neural activity in wakefulness and deep sleep, we identify the dominant interactions between neurons in each brain state. We find that during wakefulness, dominant functional interactions are pairwise while during sleep, interactions are population-wide. In particular, inhibitory cells are shown to be strongly tuned to the inhibitory population. This shows that Maximum Entropy models can be useful to analyze data-sets with excitatory and inhibitory cells, and can reveal the role of inhibitory cells in organizing coherent dynamics in cerebral cortex.