A real connectome.
A life that continues.
IMMORTAL FRUIT FLY loads the neuron list and connectivity matrix published with the Drosophila brain model by Shiu and colleagues. A worker simulates spikes over that measured network while the interface lets you preserve and restore its complete numerical state.
The data that actually runs
The bundled author-provided FlyWire v783 model contains 138,639 neurons, 15,091,983 directed connections and 54,492,922 synapses. Every source connection is retained, including single-synapse and inhibitory connections. These are counts computed from the loaded files, rather than the larger headline atlas count of 139,255 neurons.
- Shiu model repository, pinned source version: the v783 neuron list and signed connectivity matrix.
- FlyWire annotations, pinned source version: cell types, soma positions, anchor positions and laterality.
- Machine-readable data manifest: exact counts, source URLs, SHA-256 hashes and sensory group sizes.
Root IDs remain strings, avoiding loss of precision. Compressed data files are checked against SHA-256 hashes before the simulation can start. Failed downloads stop neural operation; there is no invented graph fallback.
What is measured, and what is modeled
| Element | Source or method |
|---|---|
| Neuron identity and connectivity | Electron-microscopy-derived FlyWire data in the authors’ v783 files. All 138,639 model neurons and all connection rows participate. |
| Neuron positions | Annotated soma coordinates where available, otherwise annotated anchor coordinates. Source voxels convert to micrometres at 0.004 / 0.004 / 0.04. Fourteen unlocated cells still simulate but are omitted from spatial display. |
| Connection sign | The authors’ supplied excitatory/inhibitory assignments, based on predicted neurotransmitters. Sign is a model assumption, not a direct physiological measurement. |
| Neural dynamics | Uniform leaky integrate-and-fire neurons with exponentially decaying synaptic current, following the published Shiu model approach. This is a browser implementation, not a claim of reproducing every paper result. |
| Body motion | An illustrative mapping of DNa02 left/right steering, DNa01 forward drive, MDN reverse and DNp09 stopping. Wing animation is cosmetic. There is no validated musculoskeletal body model. |
Parameters: resting and reset potential −52 mV; threshold −45 mV; membrane time constant 20 ms; synaptic time constant 5 ms; refractory interval 2.2 ms; synaptic delay 1.8 ms; weight scale 0.275 mV per signed synapse. The integration step is 0.1 ms, with an analytical linear update between discrete spike events. Inactive neurons remain at rest without fabricated background spikes.
Stimulating identifiable sensory neurons
Each user pulse applies modeled Poisson activation at 150 Hz for one second of neural time. The first 100 ms is evaluated immediately, including when continuous motion is paused. Further integration continues when playback runs. This models an experimental activation, not a recording of a fly smelling or seeing the real world.
- Sugar GRNs: 20 source-matched neurons from the authors’ sugar-sensing example. One of its 21 older IDs is absent from the v783 list and is explicitly listed in the manifest; it is not replaced by a guessed neuron.
- ORN DA1: 126 annotated DA1 olfactory receptor neurons. This is not labeled as a generic fruit-odor detector.
- Visual L1/L2: 1,586 annotated right-side L1/L2 cells. Direct activation bypasses a retinal image-to-neuron model.
The interface reports spike counts, spiking-neuron counts, mean membrane potential, actual integrated neural time, named output rates and real root IDs from the running worker. Background rendering samples 4,930 measured neuron positions; active cells and a limited selection of their strongest actual outgoing connections are overlaid. The display sample does not reduce the simulated network.
What survives a lifetime
The complete checkpoint contains every neuron’s membrane potential and synaptic current, refractory timers, accumulated spike counts, delayed-spike queues, the active-neuron list, stimulus state, simulation tick and pseudorandom-generator state. These travel with the fly’s identity, body number, position, stimulus history and lineage.
“End life” pauses the network and archives the final checkpoint. “Restore fly” gives that same state to the next body. The neural clock continues from the checkpoint; the new body’s age restarts at zero. The interaction log is external history. Synaptic weights stay fixed: this release does not claim biological memory formation, learned plasticity, consciousness or a complete living-brain reconstruction.
Save, export and resume
“Save brain” stores the checkpoint in IndexedDB. “Export brain” downloads the current full neural state as JSON. “Import brain” verifies both the body schema and neural-state compatibility before replacing the active state. A checkpoint must match the loaded dataset and model version.
Reload restores the most recently saved checkpoint; unsaved changes are not recovered. Browser-storage failures are labeled session-only, and JSON export remains available. The former 12-unit demonstration’s localStorage archives remain untouched and cannot be passed off as full neural checkpoints. The archive fingerprint identifies JSON contents, not a blockchain transaction.
The BNB Chain direction
The immortal-fruit-fly concept comes from the owner-provided X screenshot: preserve identity, state, memory, lineage and interaction history beyond an individual body. The intended chain remains BNB Smart Chain mainnet (56). This version runs its own measured-connectome simulation in the browser; storage is local. No deployed contract, wallet transaction or on-chain persistence is claimed.
Sources and credits
- Shiu et al., A Drosophila computational brain model reveals sensorimotor processing, Nature (2024).
- Dorkenwald et al., Neuronal wiring diagram of an adult brain, Nature (2024).
- Schlegel et al., Whole-brain annotation and multi-connectome cell typing of Drosophila, Nature (2024).
The original model repository’s MIT license is bundled here. Scientific data attribution is preserved separately. The website presentation was originally inspired by fruitflydev/flycoinrh (MIT); no external project’s live neural feed, wallet or chain identity is used.
This is an independent project, not affiliated with the cited researchers, FlyWire, CZ or BNB Chain. Historical butterfly data remains separately attributed in the repository and does not drive this model.
Return to the neural lab ↗