Neural network modeling of the C. elegans brain
Date:
Abstract: The nematode worm Caenorhabditis elegans (C. elegans) has long been a model organism in neuroscience. With only 302 neurons in the hermaphrodite adult, it is able to exhibit a rich repertoire of behaviors. We have long known the full wiring diagram (connectome) of this model organism and are also able to record simultaneously the activity of all its neurons. Despite this, we are yet to have a holistic computational model that can predict the output of the whole system given a pattern of inputs. The aim of this project is to approximate the input-output mapping of the C. elegans nervous system with neural networks, and to use these models to achieve three goals: 1. Predict the future network activity of real worms given its history. 2. Predict the states or behavior of real worms given their neural activity. 3. Emulate a realistic virtual worm using simulated neural activity. To that end, we employ both conventional (RNNs) and emerging (GNNs) which we compare and optimize against the neural activity and connectivity of real C. elegans. In this talk I will present our preliminary work and results related to goal 1 of this project.