Independent neuron route counts gave no clear response prediction gain in 13987 selected worm pairs

We tested whether the number of independent internal-neuron routes improves evoked calcium response detection prediction in selected C. elegans cell pairs. We linked two adult wiring maps to the Randi wild-type atlas by cell name. The sample had 13,987 pairs and 721 detected responses. Whole stimulated cells were held out of training. Controls included endpoint degree, two-step routes, shortest paths, observation count and cell identity. Adding the mean log route count changed log loss from 0.183120 to 0.182948. The reduction was 0.000172, with a conditional 95 percent shared-cell bootstrap interval from -0.000549 to 0.000834. The result gives no clear added value under the tested model. The two anatomy maps and the pooled response atlas came from different animals. Aggregate detection, fixed-prediction uncertainty, correlated controls and incomplete neuron coverage limit inference. Code, cut tables, linked features, predictions, source hashes and checks support reproduction.

C. elegans · internal neuron routes · signal propagation · connectome · held-out prediction · null result