A peptide match count gave no stable prediction gain in fixed tests of 1,278 C. elegans response directions

We tested whether a molecular peptide map adds information about neural responses after anatomical wiring and gene expression breadth are known. We joined four anatomical exports, a 92-pair molecular map and the wild type (WT) and unc-31 signal propagation atlas. All 301 neurons with anatomical records could act as relays inside their own maps. The main sample had 1,278 directions with at least three finite trial values in each genotype. Five folds withheld source expression groups and removed their labels from both training roles. Adding a peptide-receptor match count raised Brier loss from 0.082753 to 0.084248 in WT and from 0.052261 to 0.052552 in the mutant. WT gains changed sign under two other fold assignments. All planned descriptive ranges for added gains included zero. Of 130 WT detections, 45 met only the mutant equivalence criterion, while 64 met neither mutant criterion. The test did not establish added prediction value for this count. It did not directly test biological response loss. Sparse mutant data, a weak mutant baseline, fixed models and data from different animals limit the result. The package gives the paired benchmark, channel lists, source versions and hashes, original code, fitted states, forecasts, figures and reproduction steps.

connectome · structure and function · Caenorhabditis elegans · neuropeptide · unc-31 · signal propagation · prediction · open benchmark