In a landmark achievement that pushes the boundaries of computational neuroscience, researchers have created one of the most comprehensive and biophysically realistic digital brain simulations ever built: a fully modeled mouse cortex. Powered by Supercomputer Fugaku, one of the fastest high-performance computing systems in the world, the simulation offers an unprecedented window into the dynamics of the brain. Its structure, function, and the complex interactions that drive cognition, behavior, and disease.
The initiative is a joint effort between the Allen Institute and several Japanese research institutions, including the University of Electro-Communications led by Tadashi Yamazaki, Ph.D. Their goal: to build a virtual brain that is accurate enough to model disease progression, test scientific hypotheses, and ultimately accelerate breakthroughs in neurology and psychiatry.
A digital cortex with biological fidelity
The virtual mouse cortex replicates both the anatomy and the electrophysiological behavior of the real organ. It comprises:
- Nearly 10 million neurons
- 26 billion synapses
- 86 interconnected brain regions
These are not abstract mathematical constructs but detailed, biologically informed models. The structure and electrical properties of each simulated neuron are drawn from extensive datasets, including the Allen Cell Types Database and the Allen Connectivity Atlas. With these, researchers created a blueprint for how cells communicate, how signals propagate, and how networks synchronize, or fail to.
To convert these datasets into a functioning virtual brain, the team used the Brain Modeling ToolKit and the Neulite neuron simulator, enabling each digital neuron to fire, signal, and adapt in realistic patterns.
Why Fugaku makes the impossible possible
Fugaku’s architecture, 158,976 compute nodes delivering over 400 quadrillion operations per second, empowers simulations previously thought unattainable. For comparison, counting to 400 quadrillion at one number per second would take longer than the age of the universe. This computational muscle lets scientists run full-scale neural simulations that capture the flow of electrical activity across networks, allowing them to observe how:
- epileptic seizures emerge and spread
- Alzheimer-like degeneration disrupts connectivity
- neural oscillations shape attention, memory, and perception
Before this, such questions required painstaking, one-experiment-at-a-time studies on living tissue. Now they can be tested virtually, safely, and at massive scale.
A new frontier for clinical and translational research
Neurological diseases are notoriously difficult to study because damage evolves over time and often can’t be observed directly in the living human brain. With digital simulations, researchers can fast-forward through disease stages, test “what-if” scenarios, and explore how early intervention might prevent long-term decline.
“This work shows that the door is open,” says Anton Arkhipov, Ph.D., from the Allen Institute. “We can now run these simulations with accuracy and confidence. Much larger models, even human-scale simulations, are within reach.”
From mouse brain to human brain
The ambition doesn’t stop with a mouse cortex. The long-term goal is to build full-brain simulations, ultimately extending toward biophysically realistic models of the human brain. Such models could transform how clinicians understand neurological disorders, design therapies, and personalize treatment.
As Yamazaki notes: “In detailed models, every biological nuance matters. This is the path toward deeper insight into how brains function, and malfunction.” By merging deep neuroscience expertise with the raw power of supercomputing, this project marks a pivotal step forward. What was once science fiction, the idea of building a complete digital brain, is rapidly becoming a scientific reality.
