Conserved two-step routes gave no clear response prediction gain in 13987 selected worm pairs

We tested whether two-step chemical routes that keep the same intermediate cell in two adult worm maps improve evoked response predictions. We linked the Witvliet maps to the Randi wild-type atlas by cell name. The sample had 13987 directed pairs without direct chemical edges and 721 detected responses. Five folds held out stimulated left/right cell families. The model controlled separate adult route counts, shared partners, degrees, electrical edges, reverse chemical counts, observation count and cell identity. Adding conserved route count changed held-out log loss from 0.183587 to 0.183784. The reduction was -0.000196, with a conditional 95 percent cell bootstrap interval from -0.000577 to 0.000137. The interval includes zero. Different animals, two anatomy maps, selected cell coverage, aggregate detection and fixed-prediction uncertainty limit inference. This test does not rule out indirect signal flow or a small gain. Code, linked features, predictions, uncertainty draws and a figure support inspection and reproduction.

C. elegans · conserved routes · connectome agreement · evoked responses · family holdout · null result