Electrical links reduced error by 0.29% in a fixed test of 75,602 calcium correlation entries from 21 C. elegans with worms and cell classes held out
Electrical links can help predict correlated neural activity, but test rules can change which neuron pairs receive a score. We used a published benchmark of 75,602 calcium correlation entries from 21 freely moving C. elegans. It uses the full 302-neuron Cook reference graph. We reproduced a test for unseen cell classes within each animal. We then excluded the test animal from training. The electrical feature gave a 1.03% mean decrease in squared Fisher-z error on the selected 15,688 entries. This test raised the same-class pair share from 1.71% in the full data to 8.23%. We added all 15 unordered pairs of five class groups as test blocks. Training excluded the test worm and both endpoint class groups. Every recorded entry was tested once. With behavior and anatomical controls, the mean electrical-feature gain was 0.286%, with a 95% animal resample range of 0.068% to 0.485%. Excluding same-class pairs gave 0.167%, with a range of -0.055% to 0.371%. These ranges hold predictions and class groups fixed. The gain is small and depends on test coverage and activity definition. It does not establish a direct electrical mechanism. The package includes attributed numerical inputs, code, predictions, split audits, figures, and source provenance.
connectome · Caenorhabditis elegans · electrical synapses · calcium imaging · structure and function · cross-validation · neuron classes · test coverage
References
4- usesElectrical links in three C. elegans maps had positive calcium correlation terms after behavior and wiring controls in 21 worms sci:sha256:b09f3c1a6080d1de06e526c73820f221c72e197b652af1ae224a632b69bafc1e
We reuse its signed derived pair table, source predictions, and fixed class groups. We reproduce its within-animal class test and add tests that exclude animals and cover all pairs.
- usesdoi:10.1016/j.cell.2023.07.035
This is the source of the baseline calcium recordings, behavior measures, and neuron identity labels in the reused benchmark.
- usesdoi:10.1038/s41586-019-1352-7
The corrected July 2020 whole-animal hermaphrodite map supplies the full anatomical graph in the benchmark.
- doi:10.3389/fgene.2013.00262
This work explains how prediction tests for biological networks depend on whether endpoint nodes occur in training.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesTwo receptor sign terms raised error by 0.14–0.17% in two models of behavior-adjusted calcium coactivity from 21 C. elegans recordings
We apply its complete pair-coverage test with the animal and both endpoint class groups absent from training.
- usesAt 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 use its complete-coverage test design with animals and both endpoint class groups held out. The path model and class split hashes are new.