Current project

Active, Fall 2026

The Fly Brain Project

Do whole-brain fly simulations behave like a real brain on the inside?

Scientists have mapped every neuron and connection in a fruit fly’s brain, and several groups have turned that map into simulations they say behave like a fly. Nobody has checked whether the activity inside those simulations matches what is measured inside real fly brains. We are building that check and publishing the result either way.

Fig. 1 The adult fly brain, dorsal view, drawn from the wiring map: central brain, two optic lobes, antennal lobes below. The signals are simulated.
140,000
neurons in the adult fly brain
~8,400
distinct cell types
130M
synaptic contact points
10 weeks
from kickoff to preprint

Why now

  1. A brain, fully mapped

    In 2024 the FlyWire consortium finished tracing an adult fly brain: 140,000 neurons and 130 million points of contact, published as a graph.

  2. Models that act like a fly

    Several groups have turned that map into running simulations. Stimulate the taste neurons and the feeding circuit switches on. One virtual fly walks and grooms.

  3. Checked from the outside only

    Every test so far looks at behaviour. Real fly brains have been recorded from directly for decades. Nobody has compared the inside.

The simulators we test

  • Shiu et al.

    Nature, 2024

    Whole brain

    Primary target

  • flyvis

    Lappalainen et al., Nature 2024

    Visual system only, ~45,000 neurons

  • Eon Systems fly-brain

    2026, GPL

    Brain plus simulated body

The test

  1. Pick circuits with a known answer

    Two circuits whose activity has been measured in living flies and described precisely.

  2. Run the simulators, unchanged

    Do the measured patterns appear?

  3. Compare against shuffled wiring

    Same neurons, same number of connections each, scrambled partners. The control the field has been missing.

Two circuits with a known answer

Fig. 2 The compass. About 50 neurons in a ring. One bump of activity marks the fly’s heading and rotates as it turns.

Measured in living flies in 2015 and matched to the wiring in 2020. We test whether a single bump forms, holds without input, and moves smoothly.

Fig. 3 The smell library. 2,000 memory cells, drawn to scale. Each odour lights about 5% of them, a different set each time.

That sparseness is what makes the circuit a memory. We test what fraction lights up and how much the sets overlap.

Every outcome is a result

The shuffled-wiring control means there is no answer we would need to hide.

Real wiring shows the signature, shuffled wiring does not
The simulators capture something real about this circuit.
Both show it
The signature comes from generic statistics, not the specific wiring. Behavioural tests would never have caught this.
Neither shows it
The simulator’s uniform parameters are insufficient here, which is exactly what the 2025 theory paper predicts.
Signature appears only after tuning parameters
The tuning itself is the finding: it says which parameters the graph leaves undetermined.

What we publish

  • A preprint on bioRxiv

    Published whichever way the result comes out.

  • A public repository

    One command reproduces every figure from a clean checkout.

  • A reusable test suite

    Anyone can run it on their own fly-brain simulator.

Key references

  • Shiu et al., “A Drosophila computational brain model reveals sensorimotor processing,” Nature 2024. The model we are testing.
  • Beiran and Litwin-Kumar, “Prediction of neural activity in connectome-constrained recurrent networks,” Nature Neuroscience 2025. Why structure alone is not enough. Preprint is free on bioRxiv.
  • Seelig and Jayaraman, Nature 2015, and Turner-Evans et al., Neuron 2020. The compass, measured and then matched to wiring.

Get involved

One term, a small team, no biology background needed for most seats.

See open positionsAsk a question