Creating a virtual brain may sound like a science-fiction nightmare, but for neuroscientists in Japan and at Seattle’s Allen Institute, it’s a big step toward a long-held dream.

They say their mouse-cortex simulation, run on one of the world’s fastest supercomputers, could eventually open the way to understanding the mechanisms behind maladies such as Alzheimer’s disease and epilepsy — and perhaps unraveling the mysteries of consciousness.

“This shows the door is open,” Allen Institute investigator Anton Arkhipov said today in a news release. “It’s a technical milestone giving us confidence that much larger models are not only possible, but achievable with precision and scale.”

Arkhipov and his colleagues describe the project in a research paper being presented this week in St. Louis during the SC25 conference on high-performance computing. The simulation models the activity of a whole mouse cortex, encompassing nearly 10 million neurons connected by 26 billion synapses.

To create the simulation, researchers fed data from the Allen Cell Types Database and the Allen Connectivity Atlas into Supercomputer Fugaku, a computing cluster developed by Fujitsu and Japan’s RIKEN Center for Computational Science. Fugaku is capable of executing more than 400 quadrillion operations per second, or 400 petaflops.

The massive data set was translated into a 3-D model using the Allen Institute’s Brain Modeling ToolKit. A simulation program called Neulite brought the data to life as virtual neurons that interact with each other like living brain cells.

Scientists ran the program in different scenarios, including an experiment that used the full-scale Fugaku configuration to model the entire mouse cortex.

“In our simulation, each neuron is modeled as a large tree of interacting compartments — hundreds of compartments per neuron,” Arkhipov said in comments emailed to GeekWire. “That is, we are capturing some sub-cellular structures and dynamics within each neuron.”

During the full-scale simulation, it took no more than 32 seconds to simulate one second of real-time activity in a living mouse brain. “This level of performance — 32 times slower than real time — is quite impressive for a system of this size and complexity,” Arkhipov said. “It is not uncommon to see a factor of thousands of times slower for such very detailed simulations (even much smaller than ours).”

The researchers acknowledge that much more work is needed to turn their simulation into a model capable of tracing the progress of a neurological disease. For example, the model doesn’t reflect brain plasticity — that is, the brain’s ability to rewire its own connections.

“If we want to mention something specific besides plasticity, then one aspect that is missing is the effects of neuromodulators, and the other is that we currently do not have a very detailed representation of sensory inputs in our whole-cortex simulations,” Arkhipov said. “For all of these, we need much more data than currently available to make much better models, although some approximations or hypotheses could be implemented and tested now that we have a working whole-cortex simulation.”

Arkhipov said the project’s long-term goal is to simulate an entire brain, not just the cortex. “There’s a distinction between whole-cortex and whole-brain,” he pointed out. “The mouse cortex (and our model of it) contains about 10 million neurons, whereas the whole mouse brain contains about 70 million neurons.”

A human-brain simulation would require an even greater leap. The human cortex alone contains not just 10 million neurons, but 21 billion.

The good news is that a sufficiently powerful supercomputer might be up to the task. “Our work shows that very detailed microscopic-level simulations of larger brains may be possible sooner than previously expected,” Arkhipov said. “The results suggest that a simulation of the whole monkey brain (such as that of a macaque monkey with 6 billion neurons) can fit on the full-scale Fugaku system.”

Arkhipov said it was important to point out that creating a brain model on a supercomputer “does not mean that such a model is complete or accurate.”

“Here we are talking about technical feasibility of simulations, and it looks like such simulations even at the scale of the monkey brain are now within reach,” he said. “But to make such simulations biologically realistic, much more experimental data production and model building work would need to happen.”

Rin Kuriyama and Kaaya Akira of the University of Electro-Communications in Tokyo are the principal authors of the paper presented at SC25, titled “Microscopic-Level Mouse Whole Cortex Simulation Composed of 9 Million Biophysical Neurons and 26 Billion Synapses on the Supercomputer Fugaku.” In addition to Arkhipov, authors from the Allen Institute include Laura Green, Beatriz Herrera and Kael Dai. The study’s other authors are Tadashi Yamazaki and Mari Iura of the University of Electro-Communications; Gilles Gouaillardet and Asako Terasawa of the Research Organization for Information Science and Technology in Hyogo, Japan; Taira Kobayashi of Yamaguchi University; and Jun Igarashi of the RIKEN Center for Computational Science.

The Allen Institute in Seattle has released the Brain Knowledge Platform, a research aid described as the most comprehensive artificial intelligence tool available for neuroscience.

The project aims to unify brain information from dozens of collaborators, species (such as humans, other primates and mice), and samples that span early development to old age, encompassing diverse data including cell types and disease indicators.

Using AI, this data has been translated into a shared scientific language or format, allowing for “apples-to-apples” comparisons across institutions and organisms to create a much larger dataset for new insights.

“Understanding the brain is not a single institute’s effort,” said Shoaib Mufti, the Allen Institute’s head of data and technology. “So you have to bring the community together in order to understand it.”

There’s an urgent need to better prevent, diagnose and treat neurological conditions. The number of people worldwide living with or dying from conditions like stroke, Alzheimer’s disease and other dementias, and meningitis has increased significantly over recent decades, according to the Institute for Health Metrics and Evaluation.

In 2021, an estimated 3.4 billion people experienced a nervous system condition, which also includes brain injuries and migraines.

To create the Brain Knowledge Platform, the Allen Institute recruited participants to voluntarily share their data. Contributors include the Allen Institute for Brain Science, the Michael J. Fox Foundation for Parkinson’s Research, teams at the University of Washington and Harvard University, the Seattle Alzheimer’s Disease Brain Cell Atlas, or SEA-AD, and others.

Amazon Web Services engineered the tool’s core computing infrastructure while Google developed AI models for the neuroscience. Funding came from the Allen Institute as well as the National Institutes of Health’s Brain Research Through Advancing Innovative Neurotechnologies Initiative, or BRAIN Initiative.

Mufti said the resource is designed to be a “discovery platform,” not a traditional research tool where a scientist has a clear idea of what they are looking for. “How you can get to the ‘aha moments’ so you find something unexpected?” he asked.

Using the platform, scientists will be able to look across diseases. Studying the differences and similarities between people diagnosed with Alzheimer’s or Parkinson’s, for example, was previously laborious to make the data comparable.

With the Brain Knowledge Platform, “you can literally line those up side by side in the tool,” said Tyler Mollenkopf, associate director of data and technology at the Allen Institute.

While much of the data comes from research animals, information gathered from human brains — including 84 postmortem donors — is also available, stripped from identifying details.

The resource is offered to scientists for free. The team hopes more organizations contribute data and they’re devising a mechanism to provide attribution to credit researchers for their information, which could encourage sharing.

Given the massive societal impact of brain diseases “a real breakthrough is needed” to better understand them, said Mufti. “Let’s bring all the information together and make it discoverable. I’m hoping that [we] can really move the ball forward in a single community.”