Automatic assignment of neuronal identities in C. elegans

Abstract

To understand how the structured activity of large collections of neurons generates complex behaviors, it is important to study identical neural circuits across animals. Such study requires the ability to automatically assign unique identities to cells based on some combination of cell shape, position, and gene expression. Currently, there exist tools that automatically identify and track cell identities through time within an individual, but the identities assigned by these tools are not guaranteed to be consistent across animals. We present several modifications of algorithms currently in use in the Leifer Lab with the goal of automatically assigning consistent identities to each neuron in C. elegans, a organism whose wiring digram is known and thought to be invariant. These modifications are designed to make use of a newly developed multicolor labelling of each neuron which captures differences in gene expression, allowing for finer disambiguation of neurons whose positions are very similar. Each of these modified algorithms is evaluated against mock data sets for which ground truth correspondences between neurons are known; of these, a single candidate algorithm far outperforms all others. We then estimate the performance of this algorithm on a real data set for which ground truth correspondences are not known, and find that it performs no better than existing algorithms that do not make use of color information. Ultimately, we conclude that assigning consistent identities across animals is far more complex than initially anticipated and may require a fundamentally different approach.