AI-BASED ANALYSIS AND PREDICTIVE MODELING OF NOSE-MOVEMENT PATTERNS DURING A GOAL-DIRECTED LOCOMOTOR TASK IN MICE

Recently, uninstructed facial movements in mice were shown to be strongly correlated with sensory information, behavioral task variables and cognitive states. However, how they are related to instructed motor actions during goal-directed behaviors remains unclear. In this study, in order to examine whether uninstructed nose movements are related to instructed locomotor behaviors, we analyzed the occurrence of different motion syllables around the nasal area and their transitions during a self-paced goal-directed locomotor task in mice. Water-restricted mice were trained to initiate trials by spontaneously starting to run while head-fixed over a horizontal disc. After they ran for 1-2 seconds, a sound cue was played and, after sustaining locomotion for 1 second, the subjects were rewarded with a water drop. During the task performance, the mouse’s face was captured with a high-resolution camera from an oblique view. Video recordings were synchronized to other behavioral recordings, namely the disc rotation velocity and the lick timing. Each facial part in the video was labeled and tracked using a deep neural network and by using Keypoint-MoSeq, nasal movements were decomposed into elementary motion motifs, called syllables. We found that nasal movements began ~1 s before locomotion and were modulated at the timing of key task-events. We identified event-specific patterns of syllable occurrence around the key events indicating that different nose poses reflect different motor behaviors during the task. These results suggest that uninstructed nasal motion syllables constitute an anticipatory, event-locked component of goal-directed behavior.