Counts of paths with no shared relay gave no clear prediction gain in fixed tests of 13,135 C. elegans calcium responses

We tested whether paths with no shared internal neuron help predict calcium responses after direct neuron stimulation. We used all 302 canonical neuron positions and every positive chemical edge in the Cook map corrected in July 2020. We calculated exact relay route counts for all 90,902 nonself directions. We linked the counts to 13,135 usable directions in a pinned Randi atlas export. The full graph gave more routes than the 205-neuron recorded panel graph for 51.4% of all panel pairs. Before adjustment, more routes went with more detected responses. A logistic model controlled cell identity, measurement count, direct wiring, degrees, shortest paths and positive walk scores. Adding two route terms raised test Brier error from 0.04790632 to 0.04794006, a relative increase of 0.070%. The central 95% range for error decrease was -0.477% to +0.261% under endpoint weights with predictions fixed. Tests with electrical edges, older wiring, more measurements, and separate source or target groups gave no clear added gain. These ranges do not sample animals. Route counts reached their endpoint degree bound in 69.0% of primary directions with a positive bound. The result limits these route summaries in the fixed prediction models. It does not show that parallel routes have no biological role. The package preserves full route arrays, path and cut query code, source records, model states, predictions, figures, and repeat instructions.

connectome · Caenorhabditis elegans · signal propagation · parallel paths · neuron cuts · prediction · reproducible benchmark