Modeling C. Elegans Nervous System’s Behavior using Machine Learning Techniques

Gonçalo Leote Cardoso Mestre

Key information

Authors:

Gonçalo Leote Cardoso Mestre (Gonçalo Leote Cardoso Mestre)

Supervisors:

Luís Miguel Teixeira D'Avila Pinto da Silveira (Luís Miguel Teixeira D'Avila Pinto da Silveira); Ruxandra Georgeta Barbulescu (Ruxandra Georgeta Barbulescu)

Published in

11/30/2021

Abstract

Given its inner complexity and potential for human advancement resulting from a deeper understanding of its structure and functioning, the study of the human brain and nervous system is one of the greatest challenges in computational neuroscience. In order to understand its dynamics, insight may be gained from deeper knowledge of simpler and smaller organisms like the Caenorhabditis elegans, a common benchmark in computational neuroscience. This organism has a small nervous system with a connectome composed of less than 400 neurons and 15000 connections, which was already studied and reconstructed in multiple ways. One of these reconstructions is available in the NEURON simulator used in this work to generate data for common locomotion and movement behaviors of this worm. In this work five different artificial neural network architectures are implemented and compared in their ability to model the behavior of this small nervous system by predicting the output voltage on different neurons using only knowledge from the electric currents given as inputs to the system. Here it is shown that small models based on the Long Short-Term Memory Unit and the Gated Recurrent Unit architectures are able to replicate with high fidelity the mentioned behaviors of the system. These architectures are also able to produce a single model that can replicate the behavior of the system for both studied behaviors. This indicates that these architectures are appropriate for creating small black box models of nervous systems for different behaviors and should be tested for more complex nervous systems.