Automated long-term tracking and analysis of social behavior in groups of mice

Abstract

Social interaction in a group of animals has been a difficult area of study since behavior develops over long periods of time, requires laborious time consuming manual annotation, and suffers from subjective scoring. We present a computer vision based method for tracking multiple mice over long periods of time (days) without mixing individual identities within the group. Our system computes the trajectory of each individual and reconstructs high order statistical ethograms (e.g. relative posture, preferred locations, following, approaching, etc.). These correlates of social interaction can be used to study courtship, dominance and aggression, which may develop over the course of days and may not be observable in acute experiments. We show the applicability of our method in studying how social hierarchy develops between a group of two males and two females over the course of 5 days.

Additional details