Bayesian Clustering of Neural Spiking Activity Using a Mixture of Dynamic Poisson Factor Analyzers

Ganchao Wei, Ian Stevenson, Xiaojing Wang

Keywords: neural spike data, clustering, dynamic Poisson factor model (DPFA), mixture of finite mixtures model, Markov chain Monte Carlo(MCMC), Laplace Approximation

Abstract: Modern neural recording techniques allow neuroscientists to observe the spiking activity of many neurons simultaneously. Although previous work has illustrated how activity within and between known populations of neurons can be summarized by low-dimensional latent vectors, in many cases what determines a unique population may be unclear. Neurons differ in their anatomical location, but also, in their cell types and response properties. Moreover, multiple distinct populations may not be well described by a single low-dimensional, linear representation. To tackle these challenges, we develop a clustering method based on a mixture of dynamic Poisson factor analyzers (DPFA) model, with the number of clusters treated as an unknown parameter. To do the analysis of DPFA model, we propose a novel Markov chain Monte Carlo (MCMC) algorithm to efficiently sample its posterior distribution. Validating our proposed MCMC algorithm with simulations, we find that it can accurately recover the true clustering and latent states and is insensitive to the initial cluster assignments. We then apply the proposed mixture of DPFA model to multi-region experimental recordings, where we find that the proposed method can identify novel, reliable clusters of neurons based on their activity, and may, thus, be a useful tool for neural data analysis.

TL;DR: We develop an efficient sampling approach to do Bayesian clustering with a mixture of dynamic Poisson factor analyzers model and show how this model can be used to identify separate populations of neurons based on their spiking activity.