In this paper, we developed engineering concepts from control theory to make predictions about action selection in the brain of C. elegans. We analyze whole brain single cell resolution activity recordings from individual animals to characterize how linear- and non-linear dynamics govern neuronal network activity. Specifically, our mathematical architecture identifies and disambiguates linear neuronal dynamics from internally-generated (non-linear) control signals, which we find to be responsible for transitioning the network between different states that are associated with different behavioral outcomes. The control signals are identified in an unsupervised manner. Moreover, the approach makes suggestions about  which individual neurons in the network contribute to these control signals, hence suggesting candidates for important decision making units. This is a novel and useful modeling framework for the unbiased analysis of brain activity data; it can be applied to other complex systems, and it represents a novel contribution to the promising network control theory approach.

Fieseler C, Zimmer M, Kutz JN. Unsupervised learning of control signals and their encodings in Caenorhabditis elegans whole-brain recordings. J R Soc Interface. 2020 Dec;17(173):20200459. doi: 10.1098/rsif.2020.0459. Epub 2020 Dec 9. PMID: 33292096; PMCID: PMC7811586.

Link to the Royal Society click here