Characterizing social impairments in rat models of ASD

Social interaction is a core component of animal behavior, and the tracking and quantification of spontaneous social behavior presents several challenges in both computer vision and interpretation. We extend a recently developed technique for 3D kinematic tracking of single animals (DANNCE) to capture the 3D poses of freely interacting animals by tracking animal identity and refining keypoint labeling networks to maintain accuracy during touching and occlusion. Using this approach, we have acquired a rich dataset of interactions across pairings from autism spectrum disorder (ASD) knockout rats and their normal counterparts. We use a dynamical embedding approach to parse animal movement throughout solitary and social contexts to find behaviors or 'gestures' which are preferentially expressed in the social context, and timestamp periods of behavioral synchronization during interaction. We find that social exchanges differ between ASD and control animals, and preliminary analysis of these interactions in the Cntnap2 rat model suggests that epochs of synchronized behaviors are dominated by aggressive behaviors in ASD pairs and more canonical play-fighting behavior in wild type animals.