A lesion bench for a mapped nervous system
The roundworm C. elegans has 302 neurons, and the connections between them have been mapped. Touch it lightly near the head and it backs away. Touch it near the tail and it speeds forward. Labs in the 1980s worked out which cells carry that reflex by killing single neurons with a laser and seeing what the worm still did. This page lets you run the same kind of experiment on a small, simplified model of that circuit.
Click a cell to switch it off with the "laser". Click again to bring it back. Hover or focus a cell for what it does.
Each card sets up one experiment. Make a guess, run it, and see what the model does, then what the published work reports. A few cards have no published answer behind them here, and say so.
Touch cells. Five neurons run long fibres along the body wall and respond to a light touch. Three cover the front half (ALML, ALMR, AVM) and two the back half (PLML, PLMR). Their fields barely overlap, so a touch near the head reaches only the front group. A sixth cell, PVM, is left out here: it is reported as neither needed nor enough for the escape response.
Relays. The front touch cells are joined by gap junctions, direct electrical contacts, and AVM is coupled the same way to an interneuron called AVD. The back touch cells are coupled to PVC. Reviews describe AVD as needed for backing up after a head touch, and PVC as needed for speeding up after a tail touch.
Command cells. AVA drives backward crawling and AVB drives forward crawling, through two families of motor neurons along the body (A-type for backward, B-type for forward). AVD and AVA are the backward path; PVC and AVB are the forward one.
Brakes. Touch cells also make chemical contacts onto the other direction's interneurons, and the two command cells hold each other back. In this model those contacts are simple brakes, so a head touch also quiets the forward path. That is a simplification of wiring that is more tangled in the real worm.
Each cell is one number between 0 (quiet) and 1 (fully active) that drifts toward its input over about 0.15 seconds. A gap junction pulls two cells toward each other's level. A pushing contact adds the sender's level to the receiver; a brake subtracts it. AVB gets a small constant input, so the intact worm crawls slowly forward at rest.
A tap feeds a full-strength input to the touch cells on that end for 1.2 seconds. The verdict comes from how far the backward and forward motor neurons rise above rest. Switching a cell off holds it at zero and cuts all its contacts.
The connection strengths are picked by hand so the intact model and the well-known switch-offs behave like the published results. They are not measured values. Real neurons in this worm mostly signal with graded voltages rather than spikes, real synapse counts run into the dozens per pair, and many more cells take part. The worm on the left is a drawing whose crawl speed follows the motor neurons, not a body simulation.
So a match between the model and a lab result here only means the model was built to agree. The interesting part is the cards marked "model only", where the model is guessing.
Fuller versions of this idea exist, and this page leans on a much smaller slice than they do.
This one tries something narrower: one reflex, about a dozen cells, entirely in the browser, set up as a series of switch-off experiments that have known answers to check against.
The "what the sources say" notes on each card are summaries of what these papers and reviews are reported to describe, in this page's words. Details such as how complete each loss was are simplified.