Abstract
We present the Similarity Networks (SIMNETS) algorithm, a computationally efficient and scalable method for identifying groups of functionally related neurons within larger, simultaneously recorded ensembles. Our approach begins by independently measuring the intrinsic relationship between the activity patterns of each neuron across experimental conditions before making comparisons across neurons (instead of directly comparing firing patterns using measures such as correlations in firing rate or synchrony). This strategy estimates the intrinsic geometry of each neuron’s output space and allows us to capture the information processing properties of each neuron in a format that is easily compared between neurons. Dimensionality reduction tools are then used to map the pairwise neuron comparisons into a low-dimensional space where groupings of functionally related neurons are identified using clustering techniques. The algorithm’s computational complexity scales almost linearly with the number of neurons analyzed and makes minimal assumptions about single-unit encoding properties, making SIMNETS especially well-suited for examining large networks of neurons engaged in complex behaviors. We validate the ability of our approach to detect functional groupings using simulated data with known ground-truth as well as three datasets including ensemble activity from primate primary visual and motor cortex as well as rat hippocampal CA1 region.