Modeling functional cell types in spike train data

Daniel N Zdeblick, Eric Shea-Brown, Daniela Witten, Michael A Buice

Abstract: A major goal of computational neuroscience is to build accurate models of theactivity of neurons that can be used to interpret their function in circuits. Here, weexplore usingfunctional cell typesto refine single-cell models by grouping them intofunctionally relevant classes. Formally, we define a hierarchical generative model for celltypes, single-cell parameters, and neural responses, and then derive anexpectation-maximization algorithm with variational inference that maximizes thelikelihood of the neural recordings. We apply this “simultaneous” method to estimatecell types and fit single-cell models from simulated data, and find that it accuratelyrecovers the ground truth parameters. We then apply our approach toin vitroneuralrecordings from neurons in mouse primary visual cortex, and find that it yieldsimproved prediction of single-cell activity. We demonstrate that the discovered cell-typeclusters are well separated and generalizable, and thus amenable to interpretation. Wethen compare discovered cluster memberships with locational, morphological, andtranscriptomic data. Our findings reveal the potential to improve models of neuralresponses by explicitly allowing for shared functional properties across neurons.