Deep-Pose-Tracker: an automated behavioural analysis framework for Caenorhabditis elegans

Abstract

Tracking and analyzing animal behaviour is a crucial step in fields such as neuroscience and developmental biology. Behavioural studies in the nematode C. elegans, for example, help in understanding how organisms respond to external cues and how the specific physiological responses link to either instantaneous or learned behaviours. Although tracking behaviour through locomotion patterns and postural dynamics is routine, it becomes laborious and time-consuming when performed manually. Automation of this process is therefore crucial for accurate and fast detection and analysis. To this end, we report Deep-Pose-Tracker (DPT), a YOLO (You Only Look Once)-based model for automated pose detection of C. elegans from videos and images. The module is further utilized for several downstream analysis algorithms to quantify essential behavioural features, including locomotion speed, orientation, forward or reverse locomotion, and complex body bends such as omega turns. In addition, it includes eigenworms decomposition to represent complex posture dynamics in a low-dimensional space. The model shows reliable performance on the validation and test datasets, with high inference speed, while being user-friendly. DPT, therefore, can be a valuable toolkit for automated behavioural quantification of C. elegans under varying experimental stimuli.