WormPose

WormPose estimates 2D pose of Caenorhabditis elegans from video data to enable quantitative analysis of worm posture and behavior.

Key Features:

• Convolutional Neural Networks: Employs convolutional neural networks to analyze video frames for robust 2D pose estimation.
• Synthetic Generative Model: Uses a realistic synthetic generative model to create training images, removing the need for human-labeled training datasets.
• Occlusion and Coil Handling: Provides accurate pose estimates for self-occluded and coiled worm postures common in C. elegans behavior.
• Adaptability Across Imaging Conditions: Designed to operate under a range of imaging scenarios encountered in worm tracking studies.
• Long-duration, High-frame-rate Processing: Applicable to long recordings (approximately 10 hours) at sampling rates around 30 Hz for posture-scale analysis.
• Validation on Diverse Datasets: Validated using both synthetic and real-world recordings, including wild-type N2 and mutant C. elegans in naturalistic conditions.

Scientific Applications:

• Behavioral Phenotyping: Enables quantitative studies of genes, neurons, and behavior in C. elegans through precise posture measurement.
• Locomotion and State Analysis: Supports analysis of locomotion patterns and behavioral states such as roaming and dwelling.
• Posture-scale Dynamics: Facilitates posture-scale analysis over long-duration recordings to study dynamic behavioral changes.
• Comparative Mutant Analysis: Allows comparison of wild-type N2 and various mutants under on-food conditions using detailed pose data.

Methodology:

Trains convolutional neural networks on images generated by a realistic synthetic generative model and validates performance on synthetic and real recordings of N2 and mutant C. elegans.