PHENOTYPING THE SOCIAL BEHAVIOUR OF MICE USING COMPUTER VISION

Genetic factors in mice can shape social behaviour which can in turn affect phenotypes of cage mates through Indirect Genetic Effects (IGEs). We investigate how variation in Epha4 genotype affects the social behaviour of a mouse toward its cage mate and whether these interactions could mediate IGEs. In our experiment, pairs of mice were housed together, each consisting of an inbred strain mouse and a cage mate carrying either a wild-type or heterozygous Epha4 gene. Allogrooming, an affiliative interaction that contributes to maintaining social bonds, was used as the primary behaviour of interest. Manual scoring of behaviour is unreliable, being time consuming, error-prone and subject to interpretation across observers. We utilize an existing computer vision-based framework to estimate pose (position and trajectory) and use that to quantify social behaviour. To maintain consistent identities of multiple mice throughout the video, the method stitches short tracklets of individual mouse poses together based on identity similarity to maintain identity assignment throughout the video. We demonstrate the use of coat colour as a metric that can improve the tracklet stitching decisions by optimizing its use against identity similarity across different inbred strains of varying coat colours. From the pose data, using JAX Animal Behaviour System (JABS), we train classifiers to quantify target social behaviours across many videos. We demonstrate that this computer vision-based approach can accurately quantify allo-grooming and other social behaviours, enabling investigation of how Epha4 genotype variation influences social interactions that may mediate IGEs in cage mates.