Worm Lab
A browser laboratory for exploring a digital C. elegans worm: follow signals through 300 neurons, change individual cells and compare how the model responds.
- My role
- Product concept, experiment workflow, interface direction and AI-assisted implementation.
Read this case in MarkdownEvidence notes
Visit the live website
The problem
A moving simulation is hard to understand when the inputs, internal activity and resulting behavior are hidden. Visitors need a way to inspect the model and test their own questions.
Constraints
- Run entirely in the browser
- Make experiments repeatable from the same starting state
- Distinguish model output from biological measurements
Current stage
Live interactive prototype
A nervous system you can explore
Worm Lab grew from an OpenWorm-inspired idea: make a small organism's neural activity visible and open to experimentation. The browser model represents 300 neurons, 96 muscle cells and 3,689 directed weighted connections. A moving worm, a cell inspector and synchronized signal charts let visitors connect a stimulus with changes inside the model and its resulting movement.
First-hand accountObserve, inspect, intervene
Observe offers neural-activity, muscle-activity and body layers. Place food or apply a food or touch pulse from either side, then follow the signals. Circuit shows a selected cell and its neighbors, with incoming and outgoing connections ordered by weight. Stimulate a cell, silence it, restore it or disable a connection to explore its contribution.
First-hand accountRewind and compare
About 60 seconds of sampled history lets you revisit an earlier moment or resume from its saved state. Compare creates a control and an intervention from the same checkpoint, silences selected cells in the intervention and applies the same initial pulse. Both run for eight simulated seconds. A shared playhead lets you compare trajectories, neural turning and threshold events; a small or zero difference is also a valid result.
First-hand accountWhat I built around the model
The connectivity table and neural dynamics are adapted from the open-source WormBrain project, with attribution retained. Worm Lab adds the browser laboratory, cell and connection controls, synchronized telemetry, complete checkpoints and paired experiments. The deterministic engine advances at 30 ticks per simulated second. JavaScript modules and Canvas run the simulation locally in the browser, without API keys or a remote inference service.
First-hand accountWhat the model can explain
This is an educational model with fixed weights and simplified dynamics. Movement combines neural output, optional odor assistance and boundary correction, displayed separately. Anatomy and motion are schematic; potential uses arbitrary units and muscle activity is not measured force. The model does not learn or reproduce a complete biological organism. Its purpose is to make a computational model inspectable and experiments repeatable.
First-hand accountView the technical architecture
Conceptual workflow
- 01Food or touch stimulus
- 02Fixed weighted neural network
- 03Neural and muscle activity
- 04Movement and signal history
- 05Control and intervention comparison
Decisions
- Show the organism, circuit and signals on one simulation clock
- Restore complete checkpoints for repeatable experiments
- Expose neural steering, odor assistance and boundary correction separately
- Keep the simulation local to the browser, without API keys or a remote inference service
Evidence
- Public interactive laboratory with Observe, Circuit and Compare viewsFirst-hand account
- 300 neurons, 96 muscle cells and 3,689 directed weighted connections in the source tableFirst-hand account
- Cell stimulation and silencing, connection controls and sampled historyFirst-hand account
- Paired control and intervention runs with a shared playheadFirst-hand account
Safety boundary
- Educational model with fixed weights, not a biologically complete digital twin
- Movement combines neural output, optional odor steering and boundary correction
- Schematic anatomy and arbitrary units; activity is not a biological measurement
- No learning or claim of consciousness
Project documentation
Working vocabulary
JavaScript / Canvas / Browser ES modules / Deterministic simulation / Responsive CSS
The basis for this case.
Record reviewed: . This describes the published account, not a new check of the project's runtime.
My role, implementation details and unlinked outcomes are first-hand statements. Public sources support only the claims identified alongside them. A project status or screenshot does not independently verify every result.