Fruit fly brain simulation: a first study plan
Research record
- Question
- Can one reference brain response be reproduced before silencing, pruning, or body coupling?
- Evidence
- Reviewed the primary paper, author code, Eon integration report and runner, EPFL FlyGym code, and an independent FlyLab project on 2026-10-25. Repository revisions are linked below.
- Experiment
- Not run. No simulation repository checkout, environment installation, dataset download, neural run, or body simulation has been performed for this note.
- Results
- A source comparison and five-stage proposed protocol only. No local observations, accuracy, speed, hardware suitability, or successful reproduction is reported.
- Limitations
- Eon reports existing body controllers, hand-chosen mappings, and unvalidated internal dynamics. FlyGym 2.x and 1.x APIs differ. The independent FlyLab project is not selected as the biological baseline. Local compatibility remains untested.
- Reproduction
- Before any run, pin code and data versions, licenses, OS and toolchain, parameters, seeds, inputs and hashes, exact command, expected response and acceptance tolerance. Preserve controls, logs, failures, and output artifacts.
Sources
Scope correction · 2026-10-28
Current execution tracking has moved to the full-circuit feedback experiment. This original source-comparison plan is preserved; this editorial update does not turn the original plan into a newly executed experiment.
Progress
Sources reviewed 2026-10-25. No installation, data download, or simulation run.
| Candidate | Checked scope | Next decision |
|---|---|---|
| Shiu author code | Reference brain model and paper | Select one documented input–output example. |
| Eon fly-brain | Multiple backends; CPU backend listed | Check CPU-only dependencies. The README’s full environment targets Linux/CUDA; Mac compatibility is unverified here. |
| FlyGym / NeuroMechFly | Body simulator; 2.x API differs from 1.x | Pin a version and its matching example. |
| FruitFlyBrain / FlyLab | Independent skill-learning project | Keep exploratory; not the biological baseline. |
Eon’s embodied report uses existing body controllers and hand-chosen mappings; its authors identify unvalidated internal dynamics.
Next
Proposed tasks; none has run.
| Task | Progress | Next step |
|---|---|---|
| Baseline | Not run | Pin one input, expected output, command, and comparison tolerance. |
| Circuit silencing | Not run | Fix the target neurons and response measure before intervention. |
| Matched control | Not run | Compare targeted and random silencing with matched counts, inputs, and seeds. |
| Pruning / compression | Not run | Remove connections incrementally under a fixed response-error limit; retain failures. |
| Body coupling | Not run | Establish a separate FlyGym baseline before testing brain-to-controller mappings. |
Run log
No runs recorded. Environment, speed, and biological validity remain untested.
Update record
-
2026-10-28: Editorial: moved current execution tracking to the full-circuit feedback note and removed the stale initial-planning task entry. Original plan/evidence status retained.
-
2026-10-25: Linked the primary paper and author repository. No simulation executed.
-
2026-10-25: Added task fields and a compact progress/next/run-log layout.
-
2026-10-25: Checked primary sources and pinned repository revisions; added candidate decisions and staged experiments. No simulation executed.