Input similarity with the stimulated class excluded decreased hidden response error by 1.28% in a fixed ridge test across 92 C. elegans worms

We tested whether input similarity helps estimate one hidden calcium response when the stimulated neuron class is excluded from the input score. We reused a published benchmark from DANDI 001075 release 0.240930.1859 and the corrected Cook chemical map. The stimulated class was the source neuron and its named left or right partner, when present. The main panel had 11,637 animal-direction means from 581 events in 92 C. elegans worms. A baseline used static wiring, recording controls, global and spatial peer means, red signals, and a response score for cells that share direct input from the stimulated class. Adding one outside-input peer contrast decreased mean animal squared error from 0.517054 to 0.510449, or 1.28%. All five test groups improved. The fixed-prediction 95% animal resample range was 0.46% to 2.16%. In 199 whole-animal refits with selected penalties and groups fixed, the range was 0.32% to 5.17%; 198 gains were positive. The full-input contrast gave a 1.30% decrease against the same baseline. The stimulated class supplied a mean of 1.03% of full peer input weight and changed weights in 11.25% of event-target rows. Thus, its direct score contribution was small in this panel. This result supports a limited response-completion rule. Other cells remain observed. It does not identify a causal common input or predict before stimulation. Anatomy and responses came from different animals. The package preserves exact compact inputs, code, source records, all predictions, model states, resample draws, figures and repeat instructions.

connectome · Caenorhabditis elegans · input similarity · signal propagation · response completion · calcium imaging · animal validation