{"id":"doi:10.5281/zenodo.23025559","kind":"article","title":"Benchmarking a whole-brain connectome model of Drosophila against experimental data: diagnosing knockout-prediction failures and a candidate excitatory role for the water-taste neuron Usnea","authors":["Vichien Fugsukjit"],"author_count":1,"year":2026,"published":"2026-09-29","venue":"Zenodo (CERN European Organization for Nuclear Research)","cited_by":0,"doi":"10.5281/zenodo.23025559","arxiv_id":null,"nct_id":null,"source_url":"https://doi.org/10.5281/zenodo.23025559","archive_url":"https://space-kitty.com/papers/doi:10.5281/zenodo.23025559","updated":null,"pmid":null,"pmcid":null,"openalex_id":"W7214704904","pdf_url":null,"abstract":"Connectome-constrained whole-brain models are increasingly used to predict which neurons control behaviour, including through in-silico knockout screens. We re-implemented the leaky integrate-and-fire (LIF) model of the adult Drosophila brain (Shiu et al., 2024; FlyWire v783) and scored it against 16 experimentally established findings with pass/fail criteria fixed before simulation. The model passed 11/16 tests; activation (stimulus-to-motor-output) tests largely passed, whereas two of three knockout tests in the feeding circuit failed at the default input strength. We identify three candidate, separable causes of knockout failure: (i) operating point : Rattle's necessity appears only near threshold (+3% at strong vs -47% at near-threshold sugar input); (ii) a structural bottleneck : Roundup carries most of the excitatory drive to the proboscis motor neuron MN9 in the model (bilateral knockout -80% to -94% in female and male connectomes), whereas acute silencing of Roundup does not reduce proboscis extension to 50 mM sucrose in food-deprived flies; and (iii) neurotransmitter (NT) uncertainty : 85 of 406 feeding-circuit neurons have low-confidence or conflicting NT predictions. For the water-taste second-order neuron Usnea, whose GABA assignment is a low-confidence prediction, removing its inhibitory output (deletion: 10 to 42 Hz) or reversing its sign (10 to 73 Hz) rescued the failing water-to-MN9 test without losing other passed tests; among 44 uncertain cell types, no other re-signing raised water-to-MN9 above 15 Hz. At an approximately matched operating point on fresh random seeds, the Usnea knockout reduced MN9 activity in the re-signed model (-26%) but not in the original (+14%), consistent with fly data; connectivity analysis shows that a GABAergic Usnea acting through disinhibition cannot be excluded. A single-neuron knockout screen of all neurons active under one simulation seed for sugar (400) and sugar + bitter (402) inputs, confirmed on independent seeds with FDR control, identified the descending neuron DNge031 as the strongest feeding brake (bilateral knockout +87% to +175% MN9). In an independent male CNS connectome, the effect was present at the calibrated synaptic scale (+55% on pre-registered fresh seeds) but weakened at lower network drive, failing our pre-registered robustness criterion. In a pre-registered expanded benchmark built from published fly experiments (104 optogenetically activated cell types; 17 silencing tests; 3 grooming tests), the re-implemented model reproduced the sufficiency screen (balanced accuracy 0.88), but the Usnea re-signing did not improve the silencing tests overall (11/17 in both models; it gained the two Usnea tests and lost two others, at unmatched operating points), a second re-signing screen of 80 uncertain cell types in olfactory and navigation circuits found no candidate, and two further pre-registered rounds at closer-matched operating points left the result inconclusive. Shiu et al. (2024) reported that Usnea-specific knockdown of the neuropeptide-processing enzyme Amontillado phenocopies Usnea silencing; a pre-registered slow, neuropeptide-like excitatory rule added to Usnea avoided the side effects of full re-signing but was too weak to reproduce the Usnea silencing phenotype. We therefore interpret Usnea's positive effect as likely involving a neuropeptide co-transmitter that fast-synapse models do not capture. We provide testable predictions, notably identification of Usnea's neuropeptide and its fast transmitter, and release all code, pre-registration files and outputs.","abstract_license":"CC-BY (publisher license, via OpenAlex)","license":"cc-by","is_open_access":true,"retracted":false,"peer_reviewed":null,"categories":["Neurobiology and Insect Physiology Research"],"details":null,"provenance":"openalex","attribution":"Abstract of \"Benchmarking a whole-brain connectome model of Drosophila against experimental data: diagnosing knockout-prediction failures and a candidate excitatory role for the water-taste neuron Usnea\" by Vichien Fugsukjit, https://doi.org/10.5281/zenodo.23025559, used under its Creative Commons Attribution license (CC-BY); markup removed, otherwise unmodified.","topics":[{"slug":"connectomics","name":"Connectomics & Brain Mapping","domain_slug":"neuroscience"}],"cited_in_dossiers":[]}