Simulating a whole brain
A model built from an animal's wiring diagram and cell physics that reproduces how a real brain works does not yet exist.
Open in the interactive tree →A wiring diagram is not a working model: a simulation also needs each synapse's strength and sign, cell properties, neuromodulators, glia and the body the brain controls. The worm C. elegans has 302 neurons and its wiring was mapped in 1986, yet a complete and faithful simulation is still elusive. Larger brains multiply the problem by orders of magnitude.
As of October 2026
The fruit fly is the leading test case: FlyWire (2024) mapped 139,255 neurons and about 50 million synapses, and the male brain-and-nerve-cord map (Cell, September 2026) has 166,700 neurons. Models that combine connectome wiring with simplified neurons have reproduced some fly circuits, such as feeding and grooming (Nature, 2024), but not a whole animal's repertoire. A human brain has about 86 billion neurons and on the order of 100 trillion synapses, so a human connectome would need data in the zettabyte range (rough estimate from the petabytes needed per cubic millimeter), far beyond current imaging.
What is missing
- Whole-brain maps of larger animals (the mouse has about 70 million neurons) at far lower cost and faster imaging
- Measurement of synapse strengths, neuron types and neuromodulation, not just wiring
- Biophysically realistic models and validated tests of whether a simulation behaves like the animal
- Computing and storage for about 10^14 synapses in real time
- Clarity about which details actually matter for function
Becomes possible once solved
- Virtual experiments on brain disease
- Brain-inspired computing
- Possible digital copies of minds
Open steps
- Mouse-scale wiring maps Medium AI leverageWhole-brain maps of larger animals such as the mouse (about 70 million neurons) at far lower cost and faster imaging.
- Fitting synapse and cell parameters High AI leverageMeasure or infer synapse strengths, neuron types and neuromodulation that wiring maps do not show.
- Is the model like the animal? Medium AI leverageValidated tests of whether a simulation behaves like the real animal, ideally using recordings from the same individual.
- Which details matter Medium AI leverageFind out which biological details actually matter for function, so that models can leave out the rest.
- Compute for 100 trillion synapses Medium AI leverageComputing and storage for about 10^14 synapses in real time, far beyond today's simulations.
Where AI could help
Medium AI leverage. AI speeds connectome reconstruction and fits circuit models, but synapse strengths and cell physics must be measured, and the scale is enormous.
- Automating segmentation and synapse detection so larger brains become affordable to map
- Fitting unknown synapse and neuron parameters to recorded activity with deep learning
- Surrogate models that run cell-level circuits much faster than full biophysics
- Testing which biological details matter by removing them in models
Shown so far
- In October 2024 the FlyWire consortium published in Nature the first whole adult fruit-fly brain wiring diagram of 139,255 neurons, built with AI segmentation and about 33 person-years of human proofreading. source
- In September 2024 Nature published a fly-vision model in which deep learning fitted the unknown parameters of a connectome-based network, whose predictions matched neural activity measured in 26 studies. source
- In February 2026 an arXiv preprint trained a network shaped like the whole fly connectome by reinforcement learning to control a simulated fly body in walking, turning and flight (an AI controller, not a biophysical simulation). source
Prerequisites
- How nerve cells fire1952
- Worm connectome1986The worm's 302-neuron wiring map is the first test case for simulating a whole brain
- Large-scale neural recording1990–2017
- Deep Learning2012
- Brain wiring maps2024