270 — Comprehensive machine learning analysis of Hydra behavior reveals a stable basal behavioral repertoire

Hydra vulgaris is a polyp that lives in fresh water. It is a cnidarian, which means that it is very bad at tennis. It’s also on the other side of the evolutionary wall from the rest of us bilaterans, which means both that it’s a poor emotional substitute for a puppy, and that it’s a great model for understanding fundamentals of neuroscience that transcend even a basic nervous system.

Using the Bag-of-Words style of model popularized in natural language processing tasks (where video frames are the equivalent of ‘words’ and the videos themselves are the ‘bags’), Han et al trained an algorithm to begin automatically clustering behaviors in free-behaving and artificially-induced-feeding hydras.

When we cluster behaviors in other animals, we commonly use geometric methods that rely on the rigidity of the body. But hydras are squishy friends, which means that such methods are not applicable here. Instead, it is necessary to build up a dictionary of hydra behaviors from scratch, allowing any behavior to spawn from a cylinder model of a hydra along its primary axis.

The findings suggest that hydras have a small corpus of highly conserved and stereotyped behaviors. This means that even a simple set of neurons such as the hydra’s can exhibit repeated, simple behaviors, and this is not specific to bilaterians.