In-silico neuroscience · Drosophila melanogaster

A fruit fly, simulated from the measured wiring of its nervous system.

NeuroFly runs a closed-loop model of the fruit fly: brain neurons from the FlyWire connectome, a nerve cord from the MaleCNS connectome, a modelled body and a world to act in. Every behaviour it shows can be traced back to the neurons that caused it — and every assumption is stated.

Read the methods See the evidence Windows · free · no account, no advertising
7,270brain neurons from the adult female FlyWire connectomeFAFB v783 · measured
784,219signed connections, weighted by measured synapse countspair × neuropil · measured
1,045nerve-cord neurons driving six legsMaleCNS v1.0 · measured
1 msneural time step; sensing, brain, body and feedback close at 120 Hzmodel
28automated test suites, all passing for release 2.0.0verification
The NeuroFly workspace: the fly in her terrarium in the centre, the 7,270 simulated neurons at their measured positions on the right, live firing-rate traces below, and on the right an explanation of her last action
Figure 1. NeuroFly 2.0. Centre: the fly in her terrarium. Right: the simulated brain at the neurons’ measured positions, and the causal chain behind her most recent action — here a takeoff, traced from the looming stimulus through the visual looming detectors LC4/LPLC2 to the giant fiber (DNp01). Bottom: firing rates of identified neuron populations.
What NeuroFly is

Measured where it can be. Modelled where it must be. Labelled everywhere.

A connectome records which neuron connects to which, and how strongly. It does not record how the cells behave in time. NeuroFly keeps the two apart: the wiring is used exactly as published; everything the connectome does not contain is an explicit, documented model component.

Measured

Anatomy from electron microscopy

Neurons, their positions, cell types and synaptic connections come from peer-reviewed connectome releases. Nothing is tuned by hand to make a behaviour look right.

Modelled

Dynamics, senses and body as stated models

Leaky integrate-and-fire neurons, sensory transduction, a jointed body and its contact with the ground. Each is documented with its parameters and its limits.

Traced

Every action has a traceable cause

When the fly takes off, grooms or feeds, NeuroFly shows which input changed, which sensory neurons it reached, and which synapses drove the deciding neurons — from recorded spikes, not from a story.

Evidence at a glance

Does it behave like a fly for the right reasons?

Classic findings from fly neuroscience, re-run as controlled in-silico experiments with the statistics shown. Where the model does not reproduce a finding, the Evidence page says so.

Looming escape

An abrupt loom triggers takeoff through the giant fiber — 50% threshold at loom intensity 0.14, 100% above 0.2, none at the weakest stimulus. Silencing LC4 cuts giant-fiber spikes by 83%, LC4 and LPLC2 together by 99.8%.

Mechanosensory specificity

Wind does not trigger escape — 0/12 takeoffs, as in real flies, which stop rather than flee in wind. Only the auditory Johnston’s-organ neurons are wired to the giant fiber; that sound of equal strength drives it (12/12) is a prediction of the wiring.

Taste decisions

Sugar drives proboscis extension; bitter vetoes it — 88% extension at sugar 0.75 alone, 0% once bitter is added (Fisher p = 0.0014), from the measured gustatory pathways.

Self-care

Dust on the antennae triggers head grooming — 100% at full dust, 0% with the JO-F mechanosensors or the DNg12 command neurons silenced (Fisher p < 0.001).

In-silico pharmacology

Inhibition acts as a dose–response on escape — scaling all inhibitory synapses suppresses takeoff from 100% to 0%, half-maximal at 1.87× normal strength. A transmitter-class experiment, not a drug model.

Honest limits

Not every finding is reproduced — habituation, associative learning, thermal preference and four of ten activation phenotypes are not yet. They are listed with their data, because they define the next model work.

All ten benchmark experiments, figures, data and the verification suite →

Use

What it is for

Research: hypotheses before the wet lab

Silence or activate any identified cell type, change transmitter-class gains, and measure the behavioural consequence in hundreds of seeded trials — a fast way to ask which cells a circuit needs before designing the genetic experiment.

Teaching: circuits you can take apart

Students see sensation become decision become movement, neuron by neuron, and can test each step themselves. Ten guided experiments ship with their literature and statistics.

Methods: a testbed for connectome models

Every run records its seeds, model version and the SHA-256 fingerprint of each data file, and exports as CSV. Results can be reproduced exactly and compared across model variants.

Towards replacement of animal experiments Vision

Our long-term aim is a digital fly accurate enough to answer some questions that today require animals — including early-stage screening of neuroactive compounds. What that requires, and how far we are, is set out openly. Vision & roadmap

Inside the application

A laboratory, not an animation

Circuit workspace: identified neuron populations with their live firing rates and buttons to silence or activate each
Virtual genetics. Silence or activate any identified population — LC4, LPLC2, the giant fiber, DNa01/02 — or any FlyWire cell type, and watch the consequence.
Experiments workspace: guided protocols with their literature references and a run button
Guided experiments. Ten protocols from the fly literature run on a separate virtual fly, with statistics and the published finding side by side.
Sentience workspace: the eight evidence-based criteria, what is known about real flies and what the model contains
Sentience criteria. The eight criteria of Birch et al. (2021), the rating for real adult flies (Gibbons et al. 2022), and what the model does and does not contain — scope, never a score.
Specimens workspace: female BANC and male MaleCNS connectomes as separately selectable anatomy sources
Two sexes, separate specimens. The female BANC and male MaleCNS connectomes, brain and nerve cord, as browsable anatomy — the basis for the next, single-specimen model.
Vision

A fly that exists only as data — and the question of what it would take for it to feel.

Complete connectomes of the adult fly brain and of its whole central nervous system have been published within the last two years. For the first time, an entire animal nervous system can be simulated on its measured wiring. NeuroFly is built to follow that path step by step: one specimen, cell-type physiology, a calibrated body, validated behaviour.

Whether a simulated nervous system could ever have experiences is an open scientific question. We treat it as one: with published criteria, testable components and no claims beyond the evidence.

Read the vision and roadmap → · Our ethical commitments →

Get NeuroFly

Release 2.0 for Windows

Free, without advertising, for Windows 10 and 11. Because the FlyWire brain data carries a non-commercial licence, NeuroFly is and remains free of charge.

Windows installer

Installs NeuroFly with a Start-menu entry.

In preparation

Portable build

Unzip and run; nothing is installed.

In preparation

Be notified

Researchers and educators who would like early access or a release notice can write to us.

Contact

Data

Built on published connectomes

NeuroFly uses public research data under its original licences, with attribution. SHA-256 fingerprints of every data file are recorded in each run and listed on the Evidence page.

DatasetSpecimenUsed forLicence
FlyWire FAFB v783Dorkenwald et al. 2024; Schlegel et al. 2024, Natureadult female, brainbrain circuit of the running modelCC BY-NC 4.0
MaleCNS v1.0Berg et al. 2025, bioRxiv; FlyEM, HHMI Janeliaadult male, brain + nerve cordlocomotor nerve cord of the running model; anatomy explorerCC BY 4.0
BANC v888Bates et al. 2026, Natureadult female, brain + nerve cordanatomy explorer (not yet in the running model)CC BY 4.0