Modeling neural connectivity of caenorhabditis elegans

Abstract

Little research has been done regarding the use of modeling techniques to estimate the connectivity of a nervous system by analyzing data recorded from its neurons. In this thesis, three main methods were implemented to solve this task: Bayesian inference, artificial neural networks, and ODE reverse engineering. The goal is to apply these methods to data recorded from C. elegans neurons and estimate the connections between these. The top two performing methods on simulated data were feed-forward neural networks and ODE reverse engineering. The worst performing methods were recurrent neural networks and Bayesian inference. Out of the methods tried, feed-forward neural network was the most robust to changes in parameters of the simulation network and noise. Furthermore, this technique is easily generalizable since it did not rely on any particular feature of the simulation network to achieve its good performance. Nevertheless, all methods are easily reproducible for any further research.