Teach the worm to find food
Can the worm's own chemosensory circuit, read by a simple linear readout, turn a changing smell into a search strategy?
Learning curve
press TrainInside the circuit
live, current readoutSaved runs
tick runs to overlay their held out curvesNo saved runs yet. Train the real circuit and a control, save both, and compare their curves here.
What is being trained
The worm crawls at constant speed. Like a real worm it cannot smell direction, only whether the odour is getting stronger or weaker over time. When it gets weaker, the AWC neurons fire; when it gets stronger, the left ASE neuron fires.
Activity spreads through the real wiring diagram to the interneurons AIY, AIZ, AIB, RIB and RIM and the command neurons AVA and AVB. A readout of those 14 rates sets how sharply the worm turns. RIA and the head motor neurons are not read: in this simple spiking model they lock into persistent firing after the first odour signal, which real, graded worm neurons do not do.
Real worms solve this with klinokinesis: turn more when things get worse, go straight when they get better. Training has to discover a readout that does the same, using only what the circuit makes of the two sensory signals.
Training plates and held out plates put the food and the start in different places.
Inspired by: Pierce-Shimomura et al. (1999) on the pirouette strategy, and Gray, Hill and Bargmann (2005) on the AIY, AIZ and RIA turning circuit.
How to read the result
Fixed: the wiring, the synapse counts, the transmitter signs and the neuron model (Shiu et al. 2024). Learned: 15 readout numbers.
A rising curve alone proves little: a trained readout can often make use of any network. The evidence is in the comparison. Train the real circuit, save it, train a rewired control, and overlay the two held out curves. If the real wiring learns faster or generalises better, the specific map carries information the controls lack.
Senses: Odour getting weaker → AWC; Odour getting stronger → ASE left. Readout neurons: AIY left, AIY right, AIZ left, AIZ right, AIB left, AIB right, RIB left, RIB right, RIM left, RIM right, AVA left, AVA right, AVB left, AVB right.
Training uses a 0.25 ms simulation step (the lab uses 0.1 ms); firing rates in these circuits differ by about 2 percent. Dataset: Cook et al. 2019 hermaphrodite connectome with Wang et al. 2024 transmitter atlas. About this dataset