A complete fly, as data.
Our aim is a digital fruit fly so faithful to the animal that research can ask it questions that today require living flies — and that the hardest question of all, whether such a system could feel, can be examined on evidence rather than opinion.
For the first time, an entire nervous system has been mapped synapse by synapse.
In 2024 the FlyWire consortium published the complete wiring diagram of an adult fly brain — about 139,000 neurons and more than 50 million synapses — and a whole-brain model built on it predicted sensorimotor pathways that were then confirmed in living flies (Dorkenwald et al. 2024; Shiu et al. 2024). Since then, complete connectomes of the brain and nerve cord of a female and a male fly have followed (Bates et al. 2026; Berg et al. 2025).
The fruit fly is also one of biology’s most important model organisms: most human disease genes have a fly counterpart, and flies are used to screen compounds and study drug action on the nervous system (Pandey & Nichols 2011). An animal that is both completely mapped and central to biomedical research is the natural first candidate for a faithful digital counterpart.
Replace, reduce, refine — starting with neuroactive compounds
The 3Rs principle asks research to replace animal experiments where possible, reduce the number of animals, and refine procedures (Russell & Burch 1959). Regulators increasingly accept non-animal methods: in the United States the FDA Modernization Act 2.0 (2022) replaced the statutory requirement for animal tests of new drugs with “nonclinical tests”, explicitly including computer models. A validated digital fly could take part of this path for questions about the nervous system.
Step 1
Target to cells
Which identified neurons carry the receptor a compound acts on — from cell-type transcriptomes mapped onto connectome cell types.
Step 2
Receptor to physiology
How receptor occupancy changes each cell’s membrane and synapses, with measured dose–response relations and their uncertainty.
Step 3
Dose to exposure
How much reaches the nervous system, and when — uptake, distribution and clearance in the fly.
Step 4
Circuit to behaviour
Simulate the whole animal and read out behaviour — escape, walking, feeding, sleep — in thousands of seeded trials.
Step 5
Blinded validation
Predict, before the experiment, the effect of compounds whose real effect on flies is published or measured by others; publish the hit rate.
Where we are today
NeuroFly 2.0 can change the strength of whole transmitter classes (excitatory, inhibitory, dopamine, serotonin, octopamine) and measure the behavioural consequence as a dose–response curve — for example, strengthening inhibitory synapses suppresses the escape response with a half-maximal effect at 1.87× (Evidence, Figure 8). It contains no model of any specific drug, dose, uptake or metabolism, and it makes no drug predictions. Steps 1 to 5 are the work ahead.
What would it take for a simulated nervous system to feel?
Research programme, not a claim.
Whether insects are sentient is now studied with explicit criteria. The framework of Birch and colleagues (2021) lists eight neural and behavioural criteria; applied to insects, adult flies meet six of them with high or very high confidence (Gibbons et al. 2022). The 2024 New York Declaration on Animal Consciousness states that there is at least a realistic possibility of conscious experience in insects.
A simulation is different from an animal: meeting a behavioural criterion in software does not show that anything is felt. But the criteria can still be examined one by one — which circuits implement them, which the model already contains, which are missing. NeuroFly’s Sentience workspace does exactly this and deliberately reports scope, never a score.
Our long-term goal is a digital fly whose functional organisation is complete enough that the scientific case for or against its sentience can be made on the same kind of evidence as for the animal itself. We pursue that goal in the open, with the precautions described on the Ethics page, and we will never present a simulation as sentient without evidence that would convince independent researchers.
From a connectome-constrained model to a validated digital fly
Each stage has a test that decides whether it is done. Stages are not dated; they are complete when the evidence is.
1 · Closed loop on measured wiring Done · 2.0
7,270 FlyWire brain neurons, 1,045 MaleCNS nerve-cord neurons, senses routed to their real receptor neurons, a body, feedback, and a fixed 120 Hz loop.
2 · Traceable causes and virtual experiments Done · 2.0
Causal chains for every action, silencing and activation of any cell type, transmitter-class pharmacology, ten guided experiments with statistics, seeded reproducibility.
3 · One animal, both sexes In progress
Brain and nerve cord from the same specimen — BANC (female) and MaleCNS (male) — instead of today’s two-animal model. The anatomy of both is already browsable in NeuroFly 2.0.
Done when: a single-specimen model runs end to end and passes the current benchmarks.
4 · Whole-brain scale Planned
From 7,270 neurons towards the complete central nervous system, which requires a faster simulation engine.
Done when: whole-brain runs in real time on a workstation with unchanged results on the benchmarks.
5 · Cell-type physiology, neuromodulation and learning Planned
Cell-type-specific membrane properties, dopamine, serotonin and octopamine as modulators, and mushroom-body learning.
Done when: habituation, associative learning and thermal preference are reproduced.
6 · Calibrated body and behaviour Planned
A biomechanical body and quantitative comparison with recorded fly behaviour.
Done when: walking, grooming and escape kinematics fall within the measured variability of real flies.
7 · Pharmacology module Planned
Receptor-level compound action, exposure over time, and dose–response benchmarking against published fly assays.
Done when: blinded predictions for held-out compounds are published with their hit rate.
8 · Independent scrutiny Planned
Preprint, public code and data pipeline, external replication of the benchmark results.
References
- Bates AS, et al. Distributed control circuits across a brain-and-cord connectome. Nature 656, 957–970 (2026). doi:10.1038/s41586-026-10735-w
- Berg S, et al. Sexual dimorphism in the complete connectome of the Drosophila male central nervous system. bioRxiv (2025). doi:10.1101/2025.10.09.680999
- Birch J. Animal sentience and the precautionary principle. Animal Sentience 2(16) (2017). doi:10.51291/2377-7478.1200
- Birch J, Burn C, Schnell A, Browning H, Crump A. Review of the evidence of sentience in cephalopod molluscs and decapod crustaceans. LSE Consulting (2021). lse.ac.uk
- Dorkenwald S, et al. Neuronal wiring diagram of an adult brain. Nature 634, 124–138 (2024). doi:10.1038/s41586-024-07558-y
- FDA Modernization Act 2.0, S.5002, 117th Congress (2022). congress.gov
- Gibbons M, Crump A, Barrett M, Sarlak S, Birch J, Chittka L. Can insects feel pain? A review of the neural and behavioural evidence. Adv Insect Physiol 63, 155–229 (2022). doi:10.1016/bs.aiip.2022.10.001
- New York Declaration on Animal Consciousness (2024). sites.google.com/nyu.edu/nydeclaration
- Pandey UB, Nichols CD. Human disease models in Drosophila melanogaster and the role of the fly in therapeutic drug discovery. Pharmacol Rev 63, 411–436 (2011). doi:10.1124/pr.110.003293
- Russell WMS, Burch RL. The Principles of Humane Experimental Technique. Methuen, London (1959).
- Shiu PK, et al. A Drosophila computational brain model reveals sensorimotor processing. Nature 634, 210–219 (2024). doi:10.1038/s41586-024-07763-9