Innexin pair terms raised gap junction prediction error for class pairs excluded from training in a fixed test of two C. elegans maps

We tested whether paired innexin expression terms improve gap junction prediction after contact area and neuron controls. We used 18 gene rows from CeNGEN L4 data. We mapped them to 173 neurons in two C. elegans nerve rings, JSH and N2U. Each test fold excluded its full expression-class pairs from the other animal's training data. Across 9,367 test pairs, the paired terms raised Brier error by 1.75% in JSH and 2.79% in N2U. Mean Brier gain was -0.001765. Its central 95% class-weight range was -0.004911 to 0.000966. These fixed-prediction ranges do not measure variation across animals. Planned changes gave no gain in both maps. The innexin score was within the range of 39 other-gene panels matched for detection count. A secondary test removed a different source fold. It used the same mean training size but allowed the tested class pairs. The paired terms then lowered error by 4.59% and 3.71%. The change between validation scopes remained positive under the class-weight check. This result limits the prediction claim for this fixed model. It does not show that innexins lack a role in wiring. One L4 and one adult animal cannot separate age from individual variation. The package has compact input facts, source hashes, code, fitted models, predictions, figures and repeat instructions.

connectome · Caenorhabditis elegans · innexin · gap junction · gene expression · membrane contact · class pair validation · animal transfer · prediction benchmark