Nine wiring terms gave no added gain over source offsets in a nested test of 1,381 C. elegans response contrasts

We tested whether a full connectome predicts source-side response differences beyond a source-class offset. We used 1,381 ordered bilateral class pairs from a fixed C. elegans signal atlas. Each pair had four response lists with at least three finite values. We transformed each value with asinh(x/0.1). The main graph retained all 302 neurons of the Cook map corrected in July 2020. Nine wiring terms described direct weights, two-step routes, and sums of walks. Five outer folds held out unordered class pairs. Inner folds selected separate ridge penalties for source offsets and wiring. Source offsets reduced mean test squared error from 0.090456 to 0.088409, or 2.26%. The descriptive 95% class-weight range was -1.46% to +6.38%. All five combined fits removed the wiring terms and gave the offset predictions. A later test that required finite wiring penalties raised error by 0.44%. Four comparison maps gave no positive added mean gain. An exact split-product calculation put 31.7% of the positive source-side product between source classes and 68.3% within them. This is a conditional calculation on fixed lists, not an estimate of biological variance. The export lacks event and animal labels, and anatomy and responses came from different animals. The result limits these models on these entries. It does not establish equal biological responses or a lack of function in asymmetric wiring. The package preserves all five neuronal maps, derived response moments, code, provenance, model states, results, figures, and repeat instructions. A full source rebuild matched every input array.

connectome · Caenorhabditis elegans · bilateral neurons · signal propagation · ridge regression · source effects · reproducible research