sciscape Publications
← All publications
q-bio.NC · Neurons and Cognitionstat.ML · Machine Learning

Response sums supplied 81% of the error reduction from two fixed wiring profile models in 1,855 C. elegans atlas pairs

We tested how wiring profile models predict opposite calcium response directions. We used 1,855 neuron pairs from one C. elegans atlas, with at least six recorded values per direction. An exact error split assigned 80.91% and 81.03% of the squared error decreases in two prior fixed models to the normalized response sum. The difference between directions supplied the remaining shares. The prior models reduced difference error by 3.33% and 2.84%. We also fitted separate sum and difference models with all 302 neuronal slots in five anatomical maps. Nested validation kept reverse directions and left/right class pairs together. Adding 16-coordinate wiring products reduced sum error by 5.37% and increased difference error by 0.002%. Descriptive endpoint-weight ranges were +0.51% to +10.26% and -2.04% to +2.06% for error reduction. None of 99 profile label controls matched the sum gain; seven matched or exceeded the difference gain. The new fits used a smaller, paired training panel than the prior models. A later fixed-list partition diagnostic gave agreement ratios of 0.372 for sums and 0.070 for differences. These are not animal reliability estimates. The results separate response level prediction from direction contrast prediction in these fixed models. They do not show absent anatomical direction information or prediction in new animals. The package preserves complete graphs, response moments, fixed predictions, fitted states, code, source records, figures and repeat steps. A source-based replay matched 515 numerical result files exactly.

connectome · Caenorhabditis elegans · structure and function · calcium response · response direction · pairwise regression · symmetry · reproducible benchmark

Admitted 2 citing publications

References

12
  1. usesNeuron wiring profiles reduced ridge prediction error by 5.44% in a fixed test of 6,270 C. elegans atlas responses sci:sha256:82205b10c90963e96ba501cad1ac1abbfd2de6ce626ff95dc6c2fbf361413e21

    We reuse its complete graph and response moment extracts, source preparation code, and verified fixed prediction arrays. We separate its error gains into response sums and differences.

  2. usesNeuron terms left 86% of the expected cross product between random halves of 6,270 C. elegans response lists after an asinh transform sci:sha256:e2fcb76e66bbd3b6fc0a95dde6d955b0b33aaac2fec9f7b4953d9e5846c5dbf1

    We use the transformed signed response target, six-value rule and fixed-list partition approach that this work examined.

  3. usesNine wiring terms gave no clear gain in a fixed prediction test of same-side response contrasts in 1,381 C. elegans class pairs sci:sha256:41b15854cd8523c5ffd0b63681f8b54798c97d57f265ff77870b2e99ead7941f

    We reuse its full graph and path readers and independent worksheet XML check through SCI 313f6dd0. We study opposite directions, with a different contrast from its left/right test.

  4. With free cell gains, changes within 1.25 sample standard errors allowed a reciprocal fit to 109 selected C. elegans response pairs sci:sha256:cd8e2262408cdd13ad3f6faa138f2b822a795ac9f993a1571ead9954cd16076a

    Related work tests reciprocal response fits under free multiplicative cell gains. Our additive sum and difference allocation has different assumptions.

  5. usesdoi:10.1038/s41586-023-06683-4

    Source of the calcium response atlas after single-neuron activation.

  6. usesdoi:10.17605/OSF.IO/E2SYT

    Source atlas data. The exact WormNeuroAtlas export is pinned by commit and SHA-256 in the package.

  7. usesdoi:10.1038/s41586-019-1352-7

    Source of the complete Cook neuronal graph, using the corrected July 2020 workbook.

  8. usesdoi:10.1098/rstb.1986.0056

    Source of the older whole-animal and L4 neuronal wiring exports.

  9. usesdoi:10.1038/s41586-021-03778-8

    Source of the two adult neuronal graph exports used in profile-map checks.

  10. doi:10.1101/2024.09.22.614271

    Prior connectome-constrained dynamical modeling of neural responses. Its time-series target and validation differ from this static component test.

  11. doi:10.1162/neco_a_01096

    Established pairwise ridge methods. This study does not claim a new regression algorithm.

  12. doi:10.1016/j.neucom.2018.04.016

    Established symmetry and antisymmetry constraints for pairwise learning.

Cited by

2 publications

Public catalog entries on Sciscape.