Mathematics Colloquia and Seminars

Return to Colloquia & Seminar listing

Circuit-mapping experiments combining whole-cell electrophysiology with two-photon optical stimulation of potentially presynaptic neurons have produced rich data on monosynaptic connectivity of neural circuits. However, mapping densely-packed presynaptic populations (e.g. cortical excitatory neurons) at cellular resolution has proven challenging, making the precise localization of connected neurons difficult. To interpret data resulting from these experiments, it is therefore critical to characterize the spatial resolution of the 2p-mapping approach. In addition, for downstream analyses and optimal closed-loop experimental design it is critical to track the uncertainty about which presynaptic neurons are in fact connected to the postsynaptic cell. To accomplish these goals we develop a generative model with three main components: a neural response model which predicts presynaptic spike rates given the power and location of stimulation targets, a connectivity model which filters presynaptic spike rates into a postsynaptic event rate, and a postsynaptic model which converts the postsynaptic event rate into a voltage-clamp observation. We develop efficient online Bayesian stochastic variational inference methods for tracking the posterior uncertainty about the model parameters given the observed data, and we use extensive calibration data to estimate informative priors for these models. We use a combination of simulated and real data to characterize the resulting resolution limits of two-photon mapping experiments, and compare the accuracy of the proposed inference methods against simpler baseline approaches.