FlyLab
A provenance-first computational pharmacology workbench. FlyLab maps typed receptor evidence onto named circuits in the adult Drosophila MaleCNS connectome and tests how strongly each model prediction depends on wiring, assumptions, and evidence quality.
Did the predicted effect actually need the wiring?
A FlyLab run starts with a compound and a free concentration. Typed pharmacological evidence is transformed into receptor-level gain changes on a named MaleCNS circuit, simulated with a deterministic rate model and a leaky integrate-and-fire model, and exported with a notebook and full provenance.
The same Python science core runs as a command-line tool, a local FastAPI application, and a zero-install browser bench on Pyodide.
- 01Typed evidence: Kd, Ki, EC50 and IC50 kept distinct, unsupported rows left missing
- 02Connectome-dependence analysis against graded null-model ensembles
- 03Ablation, specification-robustness, and global-uncertainty analyses
- 04Claim cards and a hashed artifact manifest for byte-identical reruns
Making the connectome earn its place
Predictions on the real graph are compared with degraded ensembles that preserve progressively more network information, from size and transmitter composition to degree, edge list, and matched-strength transmitter permutations. Each comparison carries a three-way verdict against a prespecified equivalence margin, and the ladder itself is validated on synthetic graphs with planted recurrent loops.
A scale ladder across nested cuts of 1k to 50k cells shows that the dependence verdict depends on the extract, with recurrence rather than size predicting which way it falls. Negative results are reported as first-class outputs.
Evidence boundary
FlyLab evaluates model behaviour, not biological efficacy or safety. It generates no live-animal results, and there is no independent circuit-level biological validation yet.