A deep learning approach to quantify aggression between competing ant colonies

A social insect colony can be viewed as a superorganism that collectively processes and responds to environmental stimuli and even other competing superorganisms, despite the limited access to information of individual workers. While the dynamic networks within superorganisms have received considerable attention, less is known about interacting superorganisms. A fascinating example of the latter can be seen in colonies of the red imported fire ant (S. invicta), one of the few highly invasive species known to harbor intraspecies aggression. Studies have suggested that inter-colony aggression can be particularly sensitive to factors such as colony size, nestmate density and environmental parameters. However, conventional aggression bioassays have typically been limited to small numbers of individuals. In this study, we assembled a deep learning pipeline to track posture sequences of individual ants and infer their behavioral ‘states’ through high-resolution video. We applied the pipeline to investigate the effects of inter-colony separation on the frequency of aggression between two competing fire ant colonies. Our results confirm that the modulation of aggression between ant societies is contextual and more nuanced than previously thought.