lihuanlz/whole-worm-platform: A physiology-anchored, full-channel whole-worm platform: building Caenorhabditis elegans in silico, block by block, from measured

Understanding how a nervous system generates behaviour requires knowing whether the biology already measured is sufficient to explain it. Whole-animal simulators of C. elegans have so far answered by fitting: synaptic weights, channel densities, or reduced oscillator units are trained until the output looks biological — which shows that the fitter works, not that the biology is understood. Here we take the opposite route: we assemble a whole-animal model of C. elegans entirely from published experimental measurements and ask how much behaviour the literature alone can produce. Every one of the 323 connectome cells (283 neurons, 7 glial cells, 32 head muscle cells, 1 remnant; 4,450 chemical and 1,219 gap-junction edges weighted by electron-microscopy contact counts) integrates 17 ionic currents whose gating constants are anchored to voltage-clamp electrophysiology, extracted with a pipeline inherited from our cardiac channel work; a newly built 96-cell body-wall muscle layer and a 24-segment viscoelastic body close the loop through measured proprioceptive coupling. No parameter is trained (Tables 4, 5). Without fitting anything, the frozen platform quantitatively reproduces the animal’s core behavioural repertoire: forward crawling at 0.222 vs. 0.21 mm s−1 and 0.62 vs. 0.58 Hz, steady backward locomotion, salt chemotaxis through both known strategies (final approach +3.46 mm), touch- and ASH-triggered escape reversals, command-state switching, and RIS-driven arrest. Three biological results fall out. First, locomotor information does not broadcast from a central pattern generator: it travels as a proprioception-coupled stadium wave through the motor-neuron chain and the muscle rows (λ ≈ 10 segments at 0.62 Hz). Second, the platform validates mechanistically: removing the TRP-4 stretch-feedback arm reproduces the exaggerated-bending phenotype of trp-4(lf) animals, and reinstating the arm rescues wild-type curvature. Third, it exposes a discovery: the strongest extrasynaptic functional edge in the whole-animal signal-propagation atlas originates from a neuron whose two marker peptides have no recorded receptor, yielding a falsifiable ligand–receptor prediction. The remaining gaps (ASH gain modulation, a residual crawl bias, the 90 unimplemented neuropeptide systems) are reported with the same bookkeeping. Because every constant is sourced, each future experimental result enters the platform as a module, not a re-fit — turning half a century of C.elegans measurement into a living, testable whole-animal model.