Chemical walk composition gave no stable gain in two ridge models of 1,342 C. elegans unc-31 responses

We tested whether chemical steps in anatomical routes help predict unc-31 calcium responses when the measured wild-type response is supplied. We used 1,342 directed C. elegans pairs with at least three finite values per strain. The Cook graph kept all 302 neuronal slots. A damped positive walk model gave expected chemical and total step counts. The baseline included cell identity, cell-specific response slopes, wiring, sampling, nonlinear response terms and total step interactions. Adding the chemical interaction reduced main ridge test error by 0.199%, but four alternate partitions gave losses. Baseline features explained 99.07% of the added predictor variance. The common penalty often missed signals added to the remaining direction. A later fit removed this overlap within each training fold and used a separate penalty. It recovered every nonzero controlled signal in three partitions. For measured responses, its gains ranged from -0.034% to +0.006% across five partitions. Its main fixed-prediction endpoint range was -0.208% to +0.071%. These are conditional atlas results, not animal confidence limits. The result limits two prediction rules. It does not identify a synaptic or peptide mechanism. Sparse mutant data, different light-train durations and missing animal IDs limit the test. The package preserves complete numerical graphs, route features, response moments, code, model states, all controlled outcomes, source hashes, figures and repeat steps. A full offline replay matched 61 numerical result files.

connectome · Caenorhabditis elegans · unc-31 · signal propagation · chemical synapse · electrical synapse · walk sensitivity · ridge regression · structure and function