A machine-learning tool to identify bistable states from calcium imaging data

Abstract

AbstractThe use of calcium imaging to map the activation of neuronsin vivoduring behavioral tasks has resulted in advancing our understanding of how nervous systems encode sensory input and generate appropriate output. In almost all of these studies, calcium imaging is used to infer spike times or probabilities since action potentials activate voltage-gated calcium channels and increase intracellular calcium levels. However, it is well known that neurons not only fire action potentials but convey information via intrinsic dynamics such as by generating bistable membrane potential states. While a number of tools for spike inference have been developed and are currently being used, no tool exists for converting calcium imaging signals to maps of cellular state in bistable neurons.Purkinje neurons (PNs), the GABA-ergic principal neurons of the cerebellum, exhibit membrane potential bistability, firing either tonically or in bursts. Several studies have implicated the role of a population code in cerebellar function, with bistability adding an extra layer of complexity to the code. In this manuscript we develop a tool, CaMLsort, which uses convolutional recurrent neural networks to classify calcium imaging traces as arising from either tonic or bursting cells. We validated the classifier using a number of different methods and we show that the tool performs well on simulated event rasters as well as real biological data that was previously not seen by the network. This tool offers a new way of analysing calcium imaging data from bistable neurons to understand how they participate in network computation and natural behaviors.