Modelling collective behavior in groups of mice housed under semi-naturalistic conditions

Social interactions are a crucial aspect of behavior in mice. Nonetheless, it is often difficult to distinguish the effects of interactions, from independent animal behavior. Distinguishing interactions from individual preferences is important to describe how information is transmitted in a horde and to predict behavioral patterns of a whole group. We combine high-throughput data collected in mice housed and location-tracked over multiple days in an ecologically-relevant environment (Eco-HAB system) with statistical inference models to learn the rules controlling the collective dynamics of groups of 10 to 15 individuals. We reproduce the distribution for the co-localization patterns, show they are stable over time, and find that the distribution of the inferred interaction strength captures the social structure among the animals. By separating interactions from individual preferences, we show that affecting neuronal plasticity in the prelimbic cortex - a brain structure crucial for processing social information and interacting with others - does not eliminate social interactions, yet make it harder to transmit information between mice.