Meeting Abstract

112.2  Monday, Jan. 7  Quantifying inter-specific variations through the automated discovery of stereotyped behaviors BERMAN, G.J.*; CHOI, D.; BIALEK, W.; SHAEVITZ, J.W.; Princeton University; Princeton University; Princeton University; Princeton University gberman@princeton.edu

In recent years, the scientific community has learned a great deal about morphological evolution through making comparisons between closely-related species, discovering, for instance, that significant physical alterations between species can occur through a potentially reversible accumulation of single nucleotide substitutions. Applying these ideas towards the evolution of behavioral traits, however, has proven much more challenging. Much of this difficulty arises as a result of our inability to quantify behavior with the same fidelity and richness that exists in the study of morphology. In this talk, I will describe the novel metrics we have developed to quantify stereotyped movements — behaviors that an animal performs frequently and with great similarly. Using the fruit flies of the Drosophila melanogaster species subgroup as model organisms, we find that it is possible to mine high-speed movies of an animal moving in a structureless environment for such behaviors. This is achieved using a novel method that draws from ideas in information theory, non-linear dynamics, and unsupervised learning. Our method creates a well-defined statistical definition of what it means for an animal to perform a stereotyped motion, allowing for the rigorous construction of new behavioral metrics. Moreover, we show that using these quantifications, it is possible to make meaningful comparisons between these species’ behaviors, thus opening the door for further insight into the interplay between genes, neurons, and behavior.