Adding a term for paths in either adult worm raised test Brier error in a fixed model of 8,008 C. elegans atlas pairs

We tested whether short paths found in an individual adult map improve response prediction beyond paths in a pooled connectome. We used two adult C. elegans maps and the Randi signal propagation atlas. The full join had 15,384 measured pairs. The main test kept 8,008 pairs with a pooled two-step route and no direct edge. Responses were detected in 425 of 7,527 pairs with a path in either adult and 18 of 481 pairs that required pooling. A fixed logistic model controlled for trial count, route count, connection type, neuron degree, reverse edges, bilateral identity, and soma distance. Five-fold tests excluded stimulated neuron classes. Adding a term for a path in either adult raised Brier error from 0.0475173 to 0.0475304. The gain was -0.0000131. Its central 95% node-resampling range was -0.0000878 to +0.0000357. The range is conditional on fixed atlas summaries and fitted predictions. It is not an animal confidence interval. Source version and stricter endpoint checks gave similar small changes. Responding-class tests gave a small positive Brier gain. The adjusted odds-ratio range was wide, so the result does not establish equal function or zero association. A separate check used four anatomical tables on 300 neuron labels. Source matrices remain external. The package preserves code, hashes, model states, predictions, losses, figures, and rebuild instructions. Independent graph and likelihood checks passed. A fresh-source rebuild matched all 28 result files and four supporting figures.

connectome · Caenorhabditis elegans · signal propagation · graph aggregation · path provenance · optogenetics · prediction benchmark