Sensory computation and decision making in C. elegans: a computational approach

In Caenorhabditis elegans (C. elegans) and in neuroscience generally, a hierarchical view of nervous systems prevails. Roughly speaking, sensory neurons encode the external environment, interneurons encode internal state and decisions, and motor neurons encode muscle activation. Here, using an integrated approach to model sensory computation and decision making in C. elegans, I show a striking phenomenon. Via the simplest modulation possible, sensitization and desensitization, sensory neurons in C. elegans can also encode the animal’s internal state. In this thesis, I present a modeling framework, and use it to implement two detailed models of sensory adaptation and decision making. In the first model I consider a decision making task, in which worms need to cross a lethal barrier in order to reach an attractant on the other side. My model captures the experimental results, and predicts a minimal set of requirements. This model‘s mechanism is reminiscent of similar top-down attention modulation motifs in mammalian cortex. In the second model, I consider a form of plasticity in which animals alternate their perception of a signal from attractive to repulsive. I show how the model encodes high and low-level behavioral states, balancing attraction and aversion, exploration and exploitation, pushing the ‘decision making’ into the sensory layer. Furthermore, this model predicts that specific sensory neurons may have the capacity to selectively control distinct motor programs. To accomplish these results, the modeling framework was designed to simulate a full sensory motor pathway and an in silico simulation arena, allowing it to reproduce experimental findings from multiple assays. Hopefully, this allows the model to be used by the C. elegans community and to be extended, bringing us closer to the larger aim of understanding distributed computation and the integrated neural control of behavior in a whole animal.