BIO

My name is Carolina Dominguez, and I am a second year undergraduate student from Miami, FL.

Examining Animal Behavior During a Two-Response Taste Detection Task Using a Machine Learning Methodology

Abstract

DeepLabCut (DLC) is a marker less pose-estimation software package that allows high-throughput animal tracking utilizing a machine learning methodology. In this software, a model is iteratively trained with multiple video recordings of the animal and a series of user defined and labeled points. The goal of this project was to investigate the movement trajectories associated with specific behavioral responses during a two-response taste detection investigating how temperature modulates NaCl sensitivity. From 12 video recordings of experimental mice, 800 frames were extracted and expertly labeled with the following points of interest: nose, head, rear, and two stimulus presentation LEDs. The resultant dataset was then used to train a model on the HiPerGator compute cluster at the University of Florida. Experimental videos were then processed with our model to extract positional and temporal data for the points of interest. The model has 90 percent accuracy in identifying the correct positions of the points of interest despite the novelty of the videos. The positional data will soon be introduced into downstream processing pipelines to extract movement trajectories associated with either the stimulus or behavioral response. These experiments will help increase our understanding of the integration of thermal and chemosensory stimuli present in foods and beverages important for meal preferences and nutrition.

Keywords: Machine Learning, Deeplabcut, Pose estimation