When
Where
Student: Marium Yousuf, Program in Applied Mathematics
Title: Inferring Effective Connectivity Graphs from Neural Replay Spike Trains
Advisors: Michael (Misha) Chertkov (University of Arizona, Program in Applied Mathematics), Jean-Marc Fellous (University of California San Diego, Institute for Neural Computation)
Location: Zoom: https://arizona.zoom.us/j/87921043036
Abstract: Hippocampal replay is characterized by the reactivation of population-wide neural spike sequences during sleep and rest. In spatial navigation tasks, these sequences correspond to trajectories encoded by hippocampal place cells, where each neuron represents a specific spatial location. Replay is believed to play a central role in memory consolidation, long-term memory retrieval, and the organization of past experiences. Replay episodes preserve the causal firing structure of neuronal networks, providing an experimentally accessible framework for studying how memories are represented and processed at the neuronal level. Despite this structure, inferring directed interactions between neurons during replay remains an ill-posed inverse problem under noisy observations and complex underlying connectivity. We study this problem using multivariate spike trains generated by simulated CA3 place cell networks, with the goal of estimating an effective connectivity graph whose directed influences recover the underlying synaptic conductance structure. We introduce a sparse maximum likelihood model for inferring directed dependencies from spike data alone. The approach is based on a two-phase prune-and-refit procedure that estimates time-lagged causal influence while promoting sparsity and interpretability in the inferred network. The proposed method is evaluated against standard approaches, including spike-counting, cross-correlation, transfer entropy, Granger causality, point-process generalized linear models, and vector autoregressive structural causal models. All methods are assessed using known ground-truth synaptic connectivity in simulated networks. Results indicate that identifying neurons involved in replay, along with the order in which replay activity propagates through the network, can provide insight into the role of place cells in spatial navigation and may offer new perspectives on learning and decision-making in large and complex environments.