Input similarity cut hidden cell response error by 3.28% in a fixed ridge test across 92 C. elegans worms
We tested whether cells with similar anatomical inputs help estimate a hidden cell response. We used a signed benchmark of neural stimulation and two-channel fluorescence in C. elegans. The main panel had 11,637 animal-direction means from 581 events in 92 worms. Each test worm was absent from fitting and tuning. Peer summaries excluded the target and its left or right partner. A ridge model added one peer response contrast, weighted by shared chemical input. The reference had cell, recording, wiring, global response, distance and red-channel controls. Mean squared error fell from 0.53534 to 0.51777, a 3.28% decrease. All five test groups improved. The central 95% whole-animal refit range was 0.22% to 10.20%, with selected penalties fixed. More distance scales retained a 2.15% gain. Removing the most influential worm and refitting with outer groups fixed retained a 1.78% gain. Actual weights gave more gain than all 199 reassignments within coarse distance groups. The red outcome gave no clear gain. The result supports anatomy-guided response completion under this fixed rule. It does not identify causal common input or predict before stimulation. The package gives code, compact inputs, source versions and hashes, predictions, model states, figures and repeat steps.
connectome · Caenorhabditis elegans · input similarity · response completion · calcium imaging · animal validation
References
6- usesWith one response hidden, other cells' fluorescence reduced error by 5.4% after wiring and red controls in 92 C. elegans worms sci:sha256:dcb8d2bbd3a817f85ff5a18149ec49b7511272c5bf926ba5faac8ece96d14d30
We reuse the signed animal-aware event table, anatomical graph facts, and MIT static feature and ridge code.
- usesdoi:10.1038/s41586-023-06683-4
This study supplied the neural stimulation and fluorescence measurements.
- usesdoi:10.48324/dandi.001075/0.240930.1859
This fixed release is the upstream source of the reused event benchmark.
- usesdoi:10.1038/s41586-019-1352-7
The corrected July 2020 anatomical map supplies the chemical input profiles.
- doi:10.1016/j.cub.2022.06.039
Prior work linked input similarity and other network features to shared neural activity.
- doi:10.1101/2024.09.22.614271
Prior work fitted an anatomy-constrained dynamical model. We checked the May 2026 version and do not compare its accuracy with this event-summary task.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesInput similarity with the stimulated class excluded decreased hidden response error by 1.28% in a fixed ridge test across 92 C. elegans worms
We reuse its exact compact input files, peer exclusions and main animal groups. We test a new decomposition that excludes the stimulated neuron class from common-input similarity.
- usesFour cells selected by input coverage gave no clear prediction gain over random or spatial selection in a ridge test across 79 C. elegans worms
We use its shared-input cosine rule. We test a separate task with several hidden responses and a small observed set.