CTN Seminar: Visual selectivity emerges from connectome and task constraints in the fruit fly visual system

The Duke CTN group welcomes Srini Turaga, PhD, group leader at HHMI's Janelia Farms, on Thursday 4/7, at 10:30a. Dr. Turaga will present "Visual selectivity emerges from connectome and task constraints in the fruit fly visual system" LIVE in 301 Bryan Research. ABSTRACT:Efforts to generate ``connectomes'', or synaptic wiring diagrams, of large neural circuits or entire nervous systems promise an abundance of data to inform mechanistic models of neural computation. While connectomic studies have shown that neural circuits typically have sparse and highly structured connectivity patterns, it is yet unknown the degree to which such structure constrains mechanistic models of neural circuit function. We ask if it is possible to predict the tuning properties of individual neurons in real neural circuits, given only measurements of neural connectivity, simple assumptions about single neuron biophysics, and a high-level understanding of the overall computational role of the circuit. We tested this hypothesis by building a connectome-constrained model of the first two stages of the optic lobe of the \textit{Drosophila} visual system. Among ensembles of networks that were trained to predict optic flow, we frequently find networks that correctly predict motion tuning in the well-known motion selective T4 and T5 neurons. Our framework for using machine learning to combine connnectomic constraints with task constraints paves the path for the systematic combination of multiple ethologically relevant task constraints with measurements and perturbations of neural activity and behavior.