Behavioural Classification in C. elegans: a Spatio-Temporal Analysis of Locomotion
In high-density nematode populations, occlusion impedes complete pose observation, hindering quantitative analysis of social behaviors. Method: We propose an unsupervised, spatiotemporal pattern-based method to automatically extract behavior units solely from single-point motion tracking data (e.g., head or tail coordinates), without requiring predefined behavioral labels. Biological interpretability is ensured through agent-based modeling and validation against expert-annotated ground-truth behaviors. Contribution/Results: The method robustly identifies biologically meaningful locomotor modes and reconstructs known behavioral taxonomies even under partial-tracking conditions. Simulated trajectories exhibit high fidelity to real motion (mean similarity > 0.85). Compared to supervised baselines, our approach significantly improves classification robustness and accuracy in cluttered, occluded environments—enabling scalable, label-free ethological analysis of collective nematode dynamics.