Deep learning approaches to investigate olfactory responses in behaviorally divergent deer mice

Deciphering olfactory cues for sex and species identification is a critical aspect for the survival of mammalian populations. Furthermore, the impact of mating systems on the response to social olfactory cues remains uncertain. In our investigations, we selected Peromyscus polionotus and P. maniculatus, two closely related species of North American deer mice characterized by divergent mating systems. Specifically, P. polionotus is a rare example of a genetically and socially monogamous species whereas P. maniculatus is a typical example of promiscuous rodents. These two sister species serve as essential study systems for exploring variations in social behavior in correlation with the divergence of mating system. During experimental trials, individuals of both sexes were systematically exposed to the scent of a conspecific or heterospecific in a full factorial design. All trials were recorded using a high-speed overhead camera for subsequent analysis. For the automated detection of 20 anatomical landmarks in our high-framerate videos, we employed DeepLabCut, a deep-learning-based software that allows the automation of precise animal tracking while requiring minimal manual input. Subsequently, we utilized keypoint-MoSeq to identify behavioral parameters and patterns. Keypoint-MoSeq uses machine learning to cluster keypoint data into behavioral patterns without human supervision. Our integrated approach facilitates a high-throughput analysis of video recordings, reducing the need for a priori knowledge of anticipated behaviors or concerns related to biases introduced by multiple observers.