Segmentation of Mouse Social Behavior Using an Unsupervised Machine Learning Approach
PhD Student in Bioinformatics Data Science
September 26, 2022 | 4:00 PM
Ammon-Pinizzotto Biopharmaceutical Innovation (BPI) Building
Conference Room 140
Animals display diverse behavioral repertoires and identification of these behaviors is critical for deciphering the relationship between vocal communication, social behavior, and the neural circuits that encode this information. Most approaches for distinguishing behaviors are hindered by user bias and the inability to partition all of the data. To overcome these limitations, we developed an unsupervised machine-learning (ML) approach that applies a Self-Organizing Map model to group similar patterns in movement and social interaction, as defined by a set of numeric features (e.g., velocity, relative orientation) for each frame of recorded video using an egocentric framework. Temporal context during training is achieved by weighting features from previous frames. We applied our computational model to a 5-hour recording of freely socializing adult male and female mice. The model projected each frame of video into a 400-node map, with each node representing a similar set of behavioral features that described the egocentric framework of a mouse. Our preliminary results suggest that our unsupervised ML approach may represent a powerful tool for segmenting dynamic multiple-animal social behavior.
Joel received his Bachelor’s degree in Computer Science with minors in Mathematics, Computational Biology and Biological Sciences from University of Delaware. He graduated with honors. He is currently a PhD student in the Bioinformatics Data Science program at UD. He is doing research in Dr. Joshua Neunuebel’s neuroscience lab. His research interest is in understanding the neurological basis of social behavior, where he uses mouse as a model organism.