Meeting Abstract
Advances in genetics, neuroscience, and computer science facilitate the use of new and non-traditional systems to study the biological basis of phenotypic diversity. Lake Malawi cichlids are promising in this regard, having undergone explosive speciation in the past ~1 million years, radiating into ~1,000 phenotypically diverse species. We develop tools for studying the neurogenetic basis of species differences in bower building behaviors in these fishes. Bower building is a sociospatial mating behavior exhibited by ~200 species, whereby reproductive cues cause males to construct crater-like “pits” or mountain-like “castles” by moving mouthfuls of sand for days or weeks at a time. We designed an automated recording system that integrates a low-cost mini computer, a high-definition RGB camera, and a depth sensor to measure bower behaviors in many aquariums simultaneously. We train a Convolutional Neural Network (CNN) to automatically classify bower construction behaviors, spawning, and feeding behaviors from video data with high accuracy in multiple individuals and species. We analyze depth data to measure temporal patterns of bower activity, and to show that bower construction is spatially repeatable across trials. By linking video and depth data together, we quantify a remarkable phenotype in pit-castle F1 hybrids, in which males express both parental behaviors independently in sequence, first digging a pit and then building a castle. Lastly, we ground truth brain single nuclei sequencing in combination with FACS as a means for investigating the neurogenetic basis species differences in bower behavior.
