Automated Assessment of Feeding Behaviors in Mice Using Machine Vision

Behavioral analysis is paramount in many biomedical studies as it very effectively can map phenotypic differences in behavior to genotypic differences across disease models. Feeding behavior specifically is a very important behavior related to many positive and negative health consequences. Current methods for assessing feeding behavior can be logistically complicated, time-consuming, and limited in their scope – such as the ability to handle social conditions. An automated method of behavior assessment through machine vision techniques would prove useful by providing continuous measurements and high-throughput analysis of mice in social contexts. Here, we develop an approach to quantify feeding behavior from video data over 4-day, multi-mouse experiments. We validate a feeding classifier by comparing its performance to human annotation, achieving an F1-beta score of 0.9231. I use the classifier to investigate and compare feeding behavior in C57BL/6J and BTBR T+ Itpr3tf/J mice strains – the BTBR strain being commonly used as a model of autism. I demonstrate the versatility of the model by providing analysis at the strain, arena, and individual animal levels. Further, I demonstrate the applicability of the method through assessment of a relevant model of disease, the hyperphagic Leprdb/J strain