One fish, two fish, win fish, lose fish: Imaging and analyzing the fighting behavior of zebrafish in 3D

Social interactions represent some of the most intriguing aspects of animal behavior, yet principled methods for quantifying the joint actions of two individuals are lacking. We detail a novel effort to measure and model social behavior in the 3D swimming dynamics of the adult zebrafish, Danio rerio. We describe a custom tracking apparatus consisting of multiple fast cameras, a large imaging volume, and a transparent interior cage to avoid reflection artifacts. We leverage advances in convolutional neural networks to develop 3D markerless bodypoint tracking of interacting fish, while maintaining organism identity. We focus on small groups, below any obvious collective limit, yet with a rich repertoire of interacting behaviors. Specifically, we examine stereotyped male-male fighting behaviors and analyze the dynamics using short segments of bodypoint configurations to identify ethological motifs directly from tracked data. We quantify longer-time dynamics as transitions between motifs, and we repeat our analysis in mutant fish with known social deficits.