Detailed wiring gave no clear gain in a fixed ridge test of behavior correlations for 135 C. elegans neurons from 21 worms

We tested whether detailed wiring helps predict a new neuron class in a new animal. We used all 302 neuronal slots in the corrected Cook hermaphrodite map. The graph had 3,671 chemical edges and 1,091 electrical pairs. We computed calcium correlations with signed velocity, ventral head angle, and pumping in 21 baseline worms. The test covered 135 neuron identities in 66 classes and 1,746 animal and neuron observations. Each outer test excluded the complete test animal and all members of the test class group. Inner class tests selected ridge penalties and profile weights. Controls used location, broad role, and connection totals. Added chemical input, chemical output, and electrical partner profiles reduced error by 0.57%. The 95% animal and class resample range was -2.54% to +3.87%, with predictions fixed. Five other class partitions gave gains from 0.22% to 4.74%. A later comparison allowed known class relatives in training. It gave gains from 2.96% to 7.17% across six partitions. This comparison answers a different prediction question. The primary result limits this unsigned feature model and these current behavior correlations. It does not exclude gains from nonlinear encoding or dynamical models. The package preserves new correlation summaries, every prediction, split assignments, source hashes, code, figures, and repeat instructions. A separate source replay exactly reproduced the primary tables.

Caenorhabditis elegans · connectome · neural activity · behavior correlation · cross validation · electrical synapses · neuron class