At gain 0.5 in one 302-neuron C. elegans model, a coupling change of norm 0.05 changed responses to 180 inputs by at most 2.2e-10
We tested whether inputs that give unique coupling coefficients also give useful precision. We used all 302 neurons and 5,099 directed neural edges from the Cook map. The model has known unit leak, known unit inputs, full steady state observation, and free directed coefficients. Its main positive weights have incoming absolute sums of 0.5. Seven inputs added to 173 atlas source labels removed exact ambiguity, as prior SCI work predicted. However, the smallest local response sensitivity was 9.37e-9. A stable incoming AVAR coefficient change of norm 0.05 changed every response to those 180 inputs by at most 2.17e-10. All 302 inputs exposed a maximum response change of 0.0386. We then compared fixed 20-input extensions. A planned row log determinant rule improved broad sensitivity. An added weak direction rule gave 52.2 times larger smallest sensitivity on the selection model and 13.1 to 374 times larger values in 47 other model settings. The rules improved different parts of the spectrum. We preserve plans, code, source hashes, all input sets, local spectra, and both stable models. These are synthetic model tests. They do not estimate worm noise, fit biological dynamics, or give a laboratory input protocol.
connectome · Caenorhabditis elegans · identifiability · precision · perturbation design · steady response · Fisher information · whole nervous system
References
6- usesWith full steady responses and known leak, seven added inputs were necessary and sufficient for generic coupling recovery in one 302-neuron C. elegans graph model sci:sha256:393c1519d457c84e98ce9cff281609f48f692d76f0b6391b25cc1a487ce8204d
We use its seven added input labels and test its stated limit that generic full rank does not ensure precision. We check our graph and input labels against its signed export.
- usesdoi:10.1038/s41586-019-1352-7
The corrected July 2020 hermaphrodite workbook gives the full neuronal support and alternate section weights.
- usesdoi:10.1038/s41586-023-06683-4
The pinned WormNeuroAtlas occurrence export gives the 173 starting source labels. No response amplitudes are used.
- usesdoi:10.17605/OSF.IO/E2SYT
This is the source signal propagation data record. We use only source-label membership from its pinned atlas export.
- supportsdoi:10.1093/bioinformatics/btaa404
This gives prior graph methods for coefficient identification and perturbation design. We do not claim these methods are new.
- supportsdoi:10.7554/eLife.110030.3
This gives a broader dynamic neural perturbation design method. Our benchmark fixes single-neuron steady unit inputs.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesTwo full response matrices at Laplace rates 0 and 1 gave unique coupling recovery for generic weights in a 302-neuron C. elegans model with known unit leak
We use its independent graph export to check all graph entries and input labels. It provides a prior precision test.
- usesAt unit angular frequency, 173 inputs allowed generic recovery of 5,099 real couplings in a fully observed C. elegans model with known unit leak
We reuse its full graph, input sets, main synthetic weights, and stable model pair. We test its stated open limit that temporal responses can add information absent from steady responses.