At gain one, a passive electrical path score reduced calcium test error by 0.41% or increased it by 0.52% across five class splits in 21 C. elegans

We tested whether indirect electrical paths help predict shared calcium activity beyond local wiring. We used all 302 neurons and 1,091 electrical links in the Cook reference graph. We removed each tested direct link from a passive model before we calculated its correlation score. The main panel had 54,645 entries from 8,168 named neuron pairs in 21 freely moving C. elegans. Training excluded the test animal and both endpoint class groups. At fixed gain one and ridge penalty 0.001, the added score reduced mean squared Fisher-z error by 0.412%. The 95% fixed-prediction animal resample range was 0.207% to 0.625%. Four later class splits gave gains of 0.131%, -0.515%, 0.259%, and 0.060%. Every split tested the same entries once. A later two-step passive control changed the added-score gain to -7.03%. Stronger regularization reduced this added error. These later checks were exploratory. The result limits one passive model and prediction rule. It does not show that electrical paths lack function. Anatomy came from other animals, and the reference graph is a composite. The package preserves code, an exact direct-edge deletion method, graph-derived features for all 45,451 pairs, attributed activity summaries, predictions, figures, and repeat instructions.

connectome · Caenorhabditis elegans · electrical synapses · passive diffusion · calcium activity · prediction · class holdout · covariance