Identifying Neural Connectivity using Bernoulli Autoregressive Partially Linear Additive Models

Abstract

Characterising the interactions between spiking neurons is central to our understanding of cognitive processes such as memory, perception and decision making. In this work, we consider the problem of inferring connectivity in the brain network from non-stationary high-dimensional spike train data. Under a binned spike count representation of these data, we propose a Bernoulli autoregressive partially linear additive (BAPLA) model to identify the effective connectivity of a population of neurons. Estimates of the model parameters are obtained using a regularised maximum likelihood estimator, where an $\ell_1$ penalty is used to find sparse and interpretable estimates of neuronal interactions. We also account for non-stationary firing rates by adding a non-parametric trend to the model and provide an inference procedure to quantify the uncertainty associated with our estimated networks of neuronal interactions. We use synthetic data to assess the performance of the BAPLA method, highlighting its ability to detect both excitatory and inhibitory interactions in various settings. Finally, we apply our method to a neural spiking dataset from the DANDI archive, where we study the interactions of neural processes in reaction to various stimulus-response type neuroscience experiments.