Assessing temporal and spatial evolution of clusters of functionally interdependent neurons using graph partitioning techniques

Abstract

This paper suggests a new approach for identifying clusters of neurons with correlated spiking activity in large-size neuronal ensembles recorded with high-density microelectrode arrays. The nonparametric approach relies on mapping the neuronal spike trains to a 'scale space' using a nested multiresolution projection. Similarity measures can be arbitrarily defined in the scale space independent of the fixed bin width classically used to assess neuronal correlation. This representation allows efficient graph partitioning techniques to be used to identify clusters of correlated firing within distinct behavioral contexts. We use a new probabilistic spectral clustering algorithm that simultaneously maximizes cluster aggregation based on similarity measures. The technique is able to efficiently identify functionally interdependent neurons regardless of the temporal scale from which rate functions are typically estimated. We report the clustering performance of the algorithm applied to a synthesized neurophysiological data set and compare it to known clustering techniques to illustrate the substantial gain in the performance.