Simulating the Living Brain
Feature Article | Neuroscience & Technology
“ A computer model built from the fruit fly's wiring diagram alone has predicted how its brain turns taste into action, and most predictions held up in the lab. What does that mean for simulating larger brains? ”
Scientists Simulate the Fruit Fly Brain From Its Wiring Alone
For decades, neuroscience has asked a question that sounded like science fiction: can a computer reproduce what a living brain does?
A fruit fly is the latest test case. In 2024, an international team led by researchers at Princeton, the University of Cambridge, the MRC Laboratory of Molecular Biology, the University of California, Berkeley and several partner labs published a computational model of the entire adult Drosophila melanogaster brain. Built from a complete wiring diagram, it predicted how sensory signals travel through the fly's circuits to produce actions such as feeding and grooming, and many of those predictions held up in the laboratory.
The fly's brain is tiny next to ours, and the model does not think, feel or learn. What it shows is narrower and still remarkable: for some behaviours, knowing who talks to whom, and whether the message excites or inhibits, is enough to say what happens next.
At a glance
Why the Fruit Fly?
Few creatures have given biology so much. Drosophila melanogaster has been a laboratory workhorse for over a century, and its nervous system is a useful compromise: complex enough to learn, remember, navigate, court and make decisions, yet small enough to be mapped completely.
An adult fly has on the order of 140,000 neurons, against roughly 86 billion in a human brain (Azevedo et al. 2009). That difference is what makes it possible to trace every cell and connection, and why researchers treat the fly as a stepping stone towards larger brains.
How big is a brain?
The FlyWire Connectome: A Complete Wiring Diagram
Every such simulation starts with a connectome, the full map of a nervous system's connections. The model builds on FlyWire, a community effort that combined electron-microscope images of an adult female brain with machine-learning segmentation and human proofreading by hundreds of researchers and volunteers. The resulting dataset (Dorkenwald et al. 2024) describes:
- 139,255 neurons
- about 50 million chemical synapses between them
- predicted neurotransmitter identities for those synapses, which indicate whether a connection is likely to excite or inhibit its target
Earlier datasets covered only part of the brain, such as the hemibrain, which traced roughly 25,000 neurons of the central region. FlyWire follows information from sensory organs through the brain to the neurons that drive movement, and that continuous path is what a simulation needs.
Building a Functional Digital Brain
A wiring diagram is not a working brain. To bring it to life, the team (Shiu et al. 2024) gave every neuron the same simple rule, the leaky integrate-and-fire model: each cell gathers incoming signals, leaks a little of them away, and fires when its total passes a threshold before resetting.
The ingredients were deliberately sparse. The model used the connectivity itself and the predicted neurotransmitter at each synapse, treating the main inhibitory transmitters such as GABA and glutamate as inhibitory. It had essentially one tuned parameter, the strength of a synapse, and no behavioural training. It is therefore a prediction from structure, not a machine-learning model fitted to data.
How the model was built
Predicting Behaviour from Brain Wiring
The researchers activated chosen sensory neurons in the simulation, in the way sugar-sensing or vibration-sensing cells respond in a living fly, and followed the spread of activity to motor outputs. The model recovered circuits for:
- Feeding, through the motor neurons that extend the proboscis
- Taste processing, including sweet, water and bitter inputs
- Antennal grooming, driven by neurons that sense vibration and touch on the antenna
Just as important, it flagged specific neurons as sufficient or necessary for these behaviours before anyone had tested them. That turns the model into a source of hypotheses, not just a description of what is already known.
Sweet, Bitter, and Shared Neural Circuits
Taste gave the clearest illustration. The simulations indicated that sugar and water pathways share many of the same downstream neurons on the way to feeding. Silencing sugar-sensing neurons even reduced the response to water, which suggests the two reward signals work together.
Bitter, an aversive signal, behaved differently. It acted as a brake, suppressing the motor neurons that drive feeding. A system built like this can refuse a bad meal quickly without the "eat" circuitry getting in the way.
Experimental Validation
A simulation is only as good as its agreement with the animal. The team tested predictions with optogenetics, in which genetically engineered neurons are switched on or off with light, and with calcium imaging, which shows when neurons are active. Of 164 predictions that could be checked, 91 per cent were consistent with the experimental results.
The model also exposed surprises. One class of taste neurons, called Ir94e and previously linked to attraction to low salt, was predicted to inhibit feeding. When activated by light, those neurons did suppress the response to a dilute sugar solution, pointing to an aversive role that had not been recognised. Other predictions failed, and those failures are informative too: in a few cases the likeliest explanation was that the predicted neurotransmitter at a synapse was wrong.
This marks a change in how neuroscience can work. Rather than only using experiments to find out what circuits do, researchers can now ask a model where to look and let experiments confirm or correct it.
Beyond the Fruit Fly: Towards Digital Brain Models
The fly study belongs to a wider effort to build computational stand-ins for nervous systems. These are often called "digital twins", though the term promises more than any current model delivers: the fly model reproduces selected sensorimotor pathways, not the full life of a fly. Researchers hope the methods will eventually help explain how mammalian brains, and ultimately human ones, process information.
The Evolution of Connectomics
Behind the fly model stands half a century of work on the science of mapping neural connections.
Milestones in connectomics
The Roundworm and OpenWorm
The first complete connectome was that of Caenorhabditis elegans, a worm with just 302 neurons, assembled by hand from electron micrographs and published in 1986 (White et al. 1986; see also Varshney et al. 2011). The OpenWorm project builds on that map, aiming to simulate the whole worm, body and behaviour included. It shows how much more than wiring a faithful simulation needs.
The Hemibrain and FlyWire
The hemibrain (Scheffer et al. 2020) was the first dense reconstruction of a large part of the fly brain, covering roughly 25,000 neurons of the central region. FlyWire then extended the work to essentially the whole adult brain, allowing signals to be traced from sensory receptors to motor outputs.
Human Connectomics: The H01 Project
The human brain is a different order of challenge. A landmark effort reconstructed one cubic millimetre of human temporal cortex (Shapson-Coe et al. 2024), about the size of a grain of rice. That sliver held some 57,000 cells and 150 million synapses, and its images occupy about 1.4 petabytes. A whole human brain, with about 86 billion neurons, remains far out of reach.
New Imaging Technologies
Electron microscopy is still the gold standard for resolving synapses, but it is slow and costly. A newer approach, LICONN (light-microscopy-based connectomics), physically expands brain tissue in a gel roughly sixteenfold and images it with ordinary light microscopes (Tavakoli et al. 2025). Because fluorescent labels can be added, it can also show molecular features, such as the difference between excitatory and inhibitory synapses, that electron microscopy does not reveal directly. It has so far been demonstrated on small volumes of mouse brain tissue.
Why a Connectome Alone Is Not Enough
Many neuroscientists caution that anatomy is only part of the story. The fly model succeeded for particular circuits, but a living brain also depends on:
What a wiring diagram leaves out
These factors help explain how identical wiring can behave differently depending on context, and why the fly model's occasional misses are expected. Understanding a brain will require modelling how its connections change over time, not just where they are.
Forecasting Neural Activity with Artificial Intelligence
Artificial intelligence now reaches into every step of the field. It helps to trace neurons through electron-microscopy images, and increasingly it predicts neural activity itself.
ZAPBench (Lueckmann et al. 2025) invites researchers to forecast the activity of tens of thousands of neurons in a larval zebrafish. Models that treat the brain as a three-dimensional "movie" performed best at short horizons, although the authors stress that substantial room for improvement remains. Foundation models, inspired by large language models, learn the statistical regularities of neural activity from large recordings; one trained on mouse visual cortex generalised to stimuli outside its training set (Wang et al. 2025).
Artificial Intelligence and Neuroscience: A Two-Way Partnership
Early artificial neural networks drew on biological brains, and the debt is now being repaid in both directions. Machine learning helps build connectomes, and connectomes help design better models. In the fly visual system, networks constrained by the real wiring have reproduced aspects of motion processing with far fewer adjustable parameters than conventional models (Lappalainen et al. 2024).
The Future of Whole-Brain Simulation
The fly result is more than a technical stunt. It shows that accurate behavioural predictions can come from connectivity plus a modest set of computational principles.
As datasets grow and methods improve, scientists will likely build functional models of progressively larger nervous systems. A human brain simulation remains a distant prospect, probably decades away if it is achievable in any meaningful sense. Each step through worms, flies, zebrafish and mice still reveals general principles of how neural circuits compute.
Conclusion
By combining the FlyWire connectome with a simple neuron model, researchers have shown that wiring and neurotransmitter identity can predict a surprising amount of a brain's behaviour. Optogenetic and behavioural experiments upheld most of the predictions, and the model pointed to neural functions nobody had suspected.
It is also a lesson in humility. The model captures selected circuits, not a fly's inner life, and its limits mark out what is left to discover: the chemistry, plasticity and context that wiring diagrams cannot show.
If the fruit fly's brain can be simulated well enough to predict and uncover real biology, virtual brains may one day help scientists test hypotheses, probe neurological disorders and study the mechanisms of intelligence in ways that were once out of reach.
Feature Article | Neuroscience & Technology
Connectomics, computational neuroscience and whole-brain models
Editorial Note
References and further study
Azevedo, Frederico A. C., Ludmila R. B. Carvalho, Lea T. Grinberg, et al. 2009. “Equal Numbers of Neuronal and Nonneuronal Cells Make the Human Brain an Isometrically Scaled-Up Primate Brain.” Journal of Comparative Neurology 513 (5): 532–41.
Dorkenwald, Sven, Arie Matsliah, Amy R. Sterling, et al. 2024. “Neuronal Wiring Diagram of an Adult Brain.” Nature 634 (8032): 124–38.
Lappalainen, Janne K., Fabian D. Tschopp, Sridhama Prakhya, et al. 2024. “Connectome-Constrained Networks Predict Neural Activity across the Fly Visual System.” Nature 634 (8036): 1132–40.
Lueckmann, Jan-Matthis, Alexander Immer, Alex Bo-Yuan Chen, et al. 2025. “ZAPBench: A Benchmark for Whole-Brain Activity Prediction in Zebrafish.” Paper presented at the International Conference on Learning Representations (ICLR), Singapore.
OpenWorm Foundation. n.d. OpenWorm. Accessed October 9, 2026.
Scheffer, Louis K., C. Shan Xu, Michal Januszewski, et al. 2020. “A Connectome and Analysis of the Adult Drosophila Central Brain.” eLife 9: e57443.
Shapson-Coe, Alexander, MichaÅ‚ Januszewski, Daniel R. Berger, et al. 2024. “A Petavoxel Fragment of Human Cerebral Cortex Reconstructed at Nanoscale Resolution.” Science 384 (6696): eadk4858.
Shiu, Philip K., Gabriella R. Sterne, Nico Spiller, et al. 2024. “A Drosophila Computational Brain Model Reveals Sensorimotor Processing.” Nature 634 (8033): 210–19.
Tavakoli, Mojtaba R., Julia Lyudchik, MichaÅ‚ Januszewski, et al. 2025. “Light-Microscopy-Based Connectomic Reconstruction of Mammalian Brain Tissue.” Nature 642 (8067): 398–410.
Varshney, Lav R., Beth L. Chen, Eric Paniagua, David H. Hall, and Dmitri B. Chklovskii. 2011. “Structural Properties of the Caenorhabditis elegans Neuronal Network.” PLoS Computational Biology 7 (2): e1001066.
Wang, Eric Y., Paul G. Fahey, Zhuokun Ding, et al. 2025. “Foundation Model of Neural Activity Predicts Response to New Stimulus Types.” Nature 640 (8058): 470–77.
White, J. G., E. Southgate, J. N. Thomson, and S. Brenner. 1986. “The Structure of the Nervous System of the Nematode Caenorhabditis elegans.” Philosophical Transactions of the Royal Society of London B 314 (1165): 1–340.
Comments