Team

Why this project

Worm behavior is one of the most tractable, widely-used systems for mapping neural circuits to behavior, and there’s a great public dataset of perturbation videos to work with — about 300 videos at ~27,000 frames each, with motor-neuron perturbations included.

What a team could build in one day

• Get the videos loading and indexed.
• Run an existing tracker (e.g. Tierpsy or DeepLabCut) on a subset.
• Extract per-frame behavioral features.
• Look for motor-neuron-perturbation effects in the basic statistics (speed, posture, frequency of reversals, etc.).

Minimum viable demo: tracking output + a handful of summary plots showing perturbation effects.

Stretch directions

• Train a new tracker or behavior classifier.
• Contrastive embeddings of behavior across perturbations.
• Unsupervised clustering of behavioral motifs.

Compute

High — the dataset is large and storage is the practical bottleneck. Plan to subset early in the day.