Tucker Lancaster, QBioS Thesis Proposal

Quantitative Biosciences Thesis Proposal
Tucker Lancaster
School of Biological Sciences
Advisor: Dr. Patrick McGrath (School of Biological Sciences)
Open to the Community
 
Automated Behavioral Profiling in Naturalistic Underwater Environments
Monday, September 19, 2022, at 10:00 am
In Person Location: EBB 4029
Zoom Link: https://gatech.zoom.us/j/75698487488?pwd=Ma0AGkWMnEmSxJ9H9bbdrjNnRy47KU.1
Meeting ID: 756 9848 7488
Passcode: thesis

Committee Members:
Dr. Jeffrey Todd Streelman (School of Biological Sciences)
Dr. Eva Dyer (School of Biomedical Engineering)
Dr. Gordon Berman (Emory University, Departments of Biology and Physics)

Abstract:
Techniques from machine-learning (ML) and computer-vision (CV) have breathed new life into the field of ethology (the study of behavior). But in our haste to adopt these tools, we risk the very pitfall that drove the ethologists of the early 20th century to distance themselves from the field of psychology; a loss of touch with natural phenomena. Current methods in computational ethology perform remarkably well in highly constrained settings, but such paradigms cannot hope to illicit the full range of animal behavior. Further, there are many situations (e.g., field work) where constraint is simply not an option. In this thesis, I aim to develop tools for profiling animal behavior in complex naturalistic environments, with a focus on aquatic settings. Using Lake Malawi Cichlids as model organisms, I show how these methods can be integrated into existing experimental paradigms to glean new insight into the cellular and molecular mechanisms of complex social behaviors. In chapter one, I discuss the implementation of hardware and software capable of automated capture, management, and analysis of multimodal behavioral data from thirty-three semi-naturalistic experimental setups in parallel. In chapter two, I present detailed designs for an economical (yet powerful) behavioral recording device comprised of a 360° network of submersible video cameras arrayed around a 650-gallon naturalistic arena. Finally, in chapter three, I discuss protocols for efficiently processing, visualizing, and analyzing 3D behavioral data using photogrammetric reconstruction of dynamic environments together with multi-animal 3D pose estimation. Important outcomes of this work include (1) the construction of functional and persistent data collection and analysis infrastructure that will continue to drive innovative interdisciplinary research in our lab, (2) the creation and dissemination of detailed designs for replicating this infrastructure in part or full,  (3) the generation of the first 3D multi-animal annotated pose estimation dataset for fish, and (4) extensions of cutting edge computational ethology protocols to include robust individual reidentification and environmental reconstruction.