Project 07Explore · circuit explorerPython · neuPrint · pandas · matplotlib
Tastekin et al. (Cell, 2026) traced how sugar taste on the fly's labellum reaches the motor neurons that extend its proboscis. This project asks a narrow question of that paper: can its named neurons and motifs be found, exactly, in the public MaleCNS v1.0 release, and where do the two disagree? Every claim is numbered, every mismatch is logged, and the structure is yours to explore.

The hosted explorer is a single self-contained HTML file built from the frozen structural run. Choose the raw, paper-associated or candidate-comparison view; filter LB3b, LB3c and other LB3 inputs, the requested candidates, the motor types, evidence badges and sides; click any node for its raw weights, separate transmitter prediction fields, side provenance and evidence status. Edge width is 0.7 + 4.3·√(count / visible max) and the minimum-count slider is display-only. Above 260 visible edges the explorer withholds arrows and keeps the full table. Effective-score mode shows an explicit "unavailable" message rather than falling back to raw counts, because no validated selection exists yet.


| Figure 6 edge | Paper | MaleCNS v1.0 | Ledger |
|---|---|---|---|
| GNG015 → MN9 | 473 | 478 | M290 |
| ANXXX462a → GNG585 | 179 | 178 | M291 |
| GNG585 → GNG120 | 359 | 350 | M292 |
| GNG108 → MN9 | 376 | 381 | M293 |
| GNG108 → MN4a | 98 | 135 | M294 |
| GNG120 → MN9 | 441 | 443 | M295 |
| GNG120 → MN4a | 89 | 98 | M296 |
| GNG472 → MN8 | 205 | 221 | M297 |
Both values retained; neuPrint and flat weights agree with each other; whole-body or whole-edge pruning alone does not explain the differences (M1275–M1282).
A written research protocol (v0.1, 6 October 2026) came first. Work then proceeded through gates: an initial validation checkpoint (identities and sources), then a structural-connectivity checkpoint (gates 2 and 3), each ending with an artifact-integrity check that re-hashes every source, input and output, scans for credentials, and verifies local links. Source downloads and raw queries carry provenance logs; earlier results are never overwritten, only superseded by new timestamped directories. The three experiments the paper suggests are drafted as configs and deliberately not run: the method gaps above would make any "replication" a guess. Anatomy is likewise marked BLOCKED (0 of 110 requested skeletons in this build) and a status image named anatomy_BLOCKED stands where a morphology render would go.
cd projects/sugar-circuit
python3.12 -m venv .venv && .venv/bin/pip install -r requirements-validation.txt -r requirements-structural.txt
.venv/bin/python scripts/04_resolve_neuron_ids.py # compares saved primary records, writes a timestamped result dir
MPLCONFIGDIR=/tmp/mpl .venv/bin/python visualization/scripts/build.py # rebuilds figures + explorer from the frozen run
.venv/bin/python visualization/scripts/validate.py
The frozen structural results (several GB of edge tables and distance arrays) and the raw data are not in the repository; the explorer, figures, build manifests, ledgers and the validation result directory are.