Entropy-constrained reconstruction of ancestral behavioral repertoires

Although animal behavior is an important target of natural selection, we know little about how behavioral traits evolve, especially on macroevolutionary scales. Unlike morphological traits, which leave a fossil record, we have no direct window into the behaviors of long-extinct species. Existing approaches typically isolate a few predefined traits rather than model full behavioral repertoires, limiting our ability to identify large-scale evolutionary structure. Our previous work attempted to resolve this limitation, finding correlated patterns of behaviors evolving across fruit fly species, but did not result in realistic reconstructions of ancestral states. In this talk, I will present a novel ancestral state reconstruction method that infers plausible behavioral repertoires by imposing an entropy constraint on our inferred distribution of behaviors. This constraint creates effective behavioral homologies in our reconstruction, finding a representation in trait space that is more interpretable and reveals conserved axes of behavioral variation. The method is general enough to be applied to other high-dimensional phenotypes (e.g., gene-expression data), providing a new framework for reconstructing evolutionary trajectories in complex trait spaces.