Information-theoretic characterization of concurrent activity of neural spike trains

Abstract

The analysis of massively parallel spike train recordings facilitates investigation of communications and synchronization in neural networks. In this work we develop and evaluate a measure of concurrent neural activity, which is based on intrinsic firing properties of the recorded neural units. An overall single neuron activity is unfolded in time and decomposed into working and non-firing state, providing a coarse, binary representation of the neurons functional state. We propose a modified measure of mutual information to reflect the degree of simultaneous activation and concurrency in neural firing patterns. The measure is shown to be sensitive to both correlations and anti-correlations, and it is normalized to attain a fixed bounded index which makes it interpretable. Finally, the measure is compared with widely used indexes of spike train correlation. The estimate of all measures is carried out in controlled experiments with synthetic Poisson spike trains and their corresponding surrogate datasets to asses its statistical significance.