The EPFL’s Blue Brain Project has been on a 20-year mission to find the recipe to recreate the mammalian brain within a computer. If the real brain could be digitally replicated, then we can perform limitless neuroscience experiments that are not technically and ethically possible on living brains. The project faced seemingly impossible neuroscience challenges because so little was known about the brain when it started in 2005, and even supercomputers were not yet fully ready to perform the trillions of calculations per second that would be needed to solve the equations to emulate brain function. The project would need to push the frontiers of computer science to enable supercomputers to build and simulate the most complex organ of the body. The project completed its mission in December 2024, based on the mouse brain and pilot builds of parts of the rat and human brain, and has now open sourced the recipe to enable the global neuroscience community to start building and simulating the brain—potentially of any species, at any age and with any disorder—to accelerate our understanding of the brain and its diseases.

To be able to create a digital copy of the brain meant using a vast number of experimentally derived brain maps to recreate the morphological shape of every type of neuron, find the right mix of different neuron types in each brain region, work out how they are all synaptically connected, and mathematically recreate the electrochemical behavior of each type of neuron and synapse. The biggest obstacle was how to start without any blueprints and with huge gaps in our knowledge. Even the number of neurons and the parts list of neuron and synapse types were missing—let alone all their morphological shapes and electrochemical behaviors, the right mix of neurons in each brain region, and the connectome that architects how neurons are connected. Not surprisingly, the project’s launch was met with echoes of “this is impossible” throughout the neuroscience community.

Professor Henry Markram, the founder of the project, had spent the previous decade trying to experimentally reverse engineer the brain and realized that no amount of mapping would ever be enough. A radically different approach would be needed to stand a chance of mapping the brain in all its detail. He hypothesized that since every parameter in any complex system depends on every other parameter—in other words, any one piece is shaped by the others—then laying down one piece of the puzzle might start revealing the missing parts, and only a few landmark pieces might be needed to infer the entire map. He had effectively turned the intractable problem of experimentally mapping every facet of the brain into a tractable problem of identifying the smallest set of pieces needed to infer all the rest. But to test this hypothesis he had to begin building with the scraps of available information.

The conclusion reached was that one can reverse engineer the brain much faster and easier than by exhaustively measuring every facet of its structure and function, if one simultaneously builds and simulates it because the dependencies can only be leveraged and tested during the forward engineering process – reverse and forward engineering will have to go iteratively hand in hand. But it would take a team of over 1,100 scientists and engineers to develop the recipe to build digital brain tissue in this way.

It is now possible to build a digital replica of brain tissue at multiple levels of detail ranging from molecular networks, single neurons and synapses, small to large brain circuits, brain regions, systems, to the whole brain— with only a fraction of the data that would be needed to prescribe every detail. “In fact, the vast majority of the brain’s architecture can be inferred,” Markram claims. The recipe today comprises 18 million lines of code and has been detailed in around 300 peer-reviewed scientific papers.

The Swiss Federal Government funded the project with around 300 MCHF over 20 years and secured the project across funding cycles by classifying it as a Swiss National Infrastructure project—on par with the construction of bridges, tunnels, and highways. The late Charles Klieber, Secretary of State at the time, said, “Switzerland can fund finding the recipe based on the mouse brain, but Europe will need to fund testing and applying the recipe to the human brain.”

Markram united 112 laboratories in Europe and around the world, more than 500 scientists and around 100 company engagements and put together a proposal to build and simulate the human brain. The Human Brain Project entered the largest competition the European Commission had ever launched, with around 200 other large-scale, long-term competing visionary projects. The Human Brain Project proposal, judged by a large jury, including several Nobel laureates, won funding of $1 billion over 10 years. Despite intense controversy after the award was made – and a field day for the media claiming it failed before it had started - the Human Brain project researchers collectively achieved the monumental step of establishing the IT infrastructures needed to gather, organize and database the experimental data on the structure and function of the human brain from around the world – an invaluable resource today.

“The vision of the Swiss Blue Brain Project was to be able to use virtual brain tissue to overcome the limitations faced in biological experiments,” says Markram. Since more biological resolution can be added as computers become more powerful, Markram also saw the opportunity to put the progress of neuroscience on the computing roadmap by leveraging the exponential growth in computing power to allow neuroscientists to delve deeper into the brain’s secrets—much like more powerful telescopes allow astronomers to peer further into the mysteries of the universe.

The EPFL kick started the project in 2005 with the 8th largest supercomputer in the world from IBM—the very first in the Blue Gene series of supercomputers. ‘‘The project was named inline with IBM branding at the time as previously seen with ‘Deep Blue’, the supercomputer that IBM built to beat Kasparov, the world champion in chess,» says James Gonzalo King, the lead computer engineer and the oldest member of the project. «The most exciting part of the project for me has been seeing the changing shape of supercomputing and adapting alongside to build and simulate such a complex biological system as the brain.» The project pushed the boundaries of the latest computer engineering and scientific computing to extremes, using all the subsequent generations in IBM’s Blue Gene series of supercomputers and in the last phase switched to a Hewlett Package Enterprise (HPE) supercomputing solution that could scale while using commodity hardware. In the closing phase of the Blue Brain Markram worked with Amazon to migrate our HPC processes into the cloud. Today all the software can run on the cloud, with the goal of scaling in size beyond the limits of a single cluster in a data center, ready to serve all researchers that want to accelerate their research with simulation neuroscience. «With providers now offering supercomputing in the cloud and neuroscience software leveraging the exponential growing computational power available, we are eager to jump on board with the scalability and elasticity of the cloud,’’ King added.

“We call this specialization of neuroscience, simulation neuroscience,” says Dr. Michael Reimann, the lead neuroscientist on the project. “Simulation neuroscience differs from traditional theoretical neuroscience that takes a top-down approach by searching for the simplest possible explanation (the essence) of how the brain works. In contrast, simulation neuroscience takes a bottom-up, highly data-driven approach, trying to understand the role of every component of the brain—even down to a specific molecule— in generating its remarkable cognitive superpowers.”

Markram’s experimental lab at EPFL was on standby to help the Blue Brain rush in and map parts of the brain in case the project hit a wall where the missing pieces could not be inferred - which would of course risk blocking progress. But until the recipe was fully deciphered, he would need many more experimental labs to help map the brain, so he developed a network of over 50 collaborators from around the world. Yet even that was not enough.

Markram triggered a global race to experimentally map the brain of various species when he founded and won the $1 billion grant for the EU’s Human Brain Project. And even though controversy raged more intensely around the feasibility of simulating the human brain, the race was on. National initiatives sprung up everywhere—in China, Japan, Australia, Canada, and the US. The US launched the biggest initiative, the Brain Initiative, with around a $6 billion commitment to map all the types of neurons—the parts list—in the brain. Canada recruited Sean Hill, Blue Brain’s architect for databasing the brain with next-generation knowledge engineering to help map the diseases of the brain for clinicians. The Allen Institute, funded by the late Paul Allen and co-founder of Microsoft ramped up their industrial approach to reverse engineering the brain, mapping one aspect of the brain after another down to which genes are switched on in different neurons across the entire brain. Janelia Research Campus, funded by the Howard Hughes Medical Institute, began mapping the complete morphological shapes of neurons, while China started mapping how the axons of neurons extended across the whole brain, connecting brain regions.

In the meantime, the Blue Brain was building the tools to absorb this deluge of data to build digital brain models according to these biological specifications and leverage the interdependencies in the data.

The Blue Brain Project built artificial intelligence search tools to automatically read the neuroscience literature and push the boundaries of knowledge engineering to register data in knowledge graphs and anchor it spatially in brain atlases—creating a unified and semi–self-updating library of the latest brain mappings. AI tools were also developed to extract parameters needed from the literature and in brain databases to feed algorithms developed to build digital brain models, creating a living, continuously updating library of the latest neuroscience discoveries around the world.

The next seemingly insurmountable challenge that the project faced as it began finalizing the recipe was how to make the data and all the digital brain-building technologies accessible to any neuroscientist in the world. Amazon’s AWS came to the rescue, generously providing a freely available academic platform for anyone to access all of Blue Brain’s data as part of their Academic Sponsorship Program. The data is now publicly available in the AWS Open Data Registry, including millions of neuronal morphologies and electrical recordings of neurons, synapses, ion channels, brain models as well as simulation experiments and around 290 repositories were created on GitHub featuring the various software packages. The brain atlas environment also serves as a single point of entry to brain databases around the world.

Early on Markram started preparing the next generation neuroscientist for simulation neuroscience by developing a series of Massive On-line Courses (MOOCs) that have been taken by over 22’000 students to date.

Markram also saw, despite publishing nearly 300 papers on building digital brain tissue, that it was hard for neuroscientists to switch to the new paradigm of simulation neuroscience and believe that it is possible to build biologically accurate digital replicas of the brain. So, the Blue Brain team started developing a unifying platform that would bring together all the steps, making it possible for others to test and try by themselves. The initial version of the platform was pre-demoed at the 2024 annual Society for Neuroscience conference in Chicago, attended by tens of thousands of neuroscientists. There, the Blue Brain team demonstrated how to launch a virtual neuroscience laboratory and build digital brain models in a series of workshops.

The Blue Brain team worked with AWS to solve HPC-in the cloud and all supercomputing can now run on AWS. The platform is also being ported to Microsoft Azure cloud infrastructure to provide large scale compute resources to researchers.

As the project approached completion—after a decade of preparation starting in the Max Planck Institute in Heidelberg and continuing at the Weizmann Institute with funding from the US Office of Naval Research, and two decades of racing to find the recipe to build the brain within a computer at the EPFL with unprecedented funding from the Swiss Federal Government—Markram faced another challenge. Most of the academic research phase would soon be over, and an operational, service phase would need to begin.

Markram formed a not-for-profit foundation—the Open Brain Institute— that launched in January 2025, with the mission to freely share the recipe to explore, build, and simulate the brain, empowering researchers and organizations to accelerate their research. The OBI has entered into an agreement with the EPFL to ensure that digital brain-building technologies remain open for academic research for decades to come. «EPFL is delighted to see that the extensive work accomplished within the EPFL Blue Brain Project (BBP) will impact and continue under Open Brain Institute, providing researchers worldwide with access to cutting-edge virtual laboratories,” said Anna Fontcuberta i Morral, EPFL’s current President.

The OBI will host virtual labs running on the open brain platform where researchers can freely use the models built by the Blue Brain, modify them, or even build their own; then plan experiments and test ideas—much more quickly and cheaply than with trial biological experiments—and start exploring aspects of brain structure and function that are impossible to probe on actual brain tissue. Models built by the community will also be freely accessible to members of other virtual labs, fueling an acceleration in digital brain building. “We hope it triggers an age of digital brains where tens of thousands of neuroscientists around the world build digital brain tissue models for different animal species, across the stages of brain development and aging, and for various brain diseases,” says Henry Markram.

Yet accessibility would still be a major challenge for even the most expert neuroscientist, because the recipe is made up of a huge number of values, equations, algorithms, processes, and workflows—all organized into a suite of software applications amounting to millions of lines of code—that would limit its dissemination. “A solution emerged. We are in the midst of an AI revolution where Large Language Models (LLMs) can capture all the world’s knowledge and serve it in any language, including the language of computers. So, we wrapped the applications in LLMs, giving them access to the recipe, simplifying interactions with the digital brain-building technologies,” says Jean-Denis Courcol, the Director of Engineering during the Blue Brain Project and now CTO of the OBI. “We are on a path that will make it possible for researchers to perform neuroscience at the speed of thought,” says Georges Khazen, one of the first of 45 PhD students of the Blue Brain Project, and now the CEO of OBI.

The OBI, currently funded by the Markram family, hired 37 of the key members of the Blue Brain Project to help the research community set up their own virtual neuroscience laboratories where they can build digital brain models to complement their research with the new third pillar of neuroscience alongside the experimental and theoretical pillars—simulation neuroscience. “Our mission is to make science accessible so that we can all live healthy and prosperous lives on a healthy planet,” says Kamila Markram, Vice President of the OBI, CEO and co-founder of Frontiers, a pioneer of open access science publishing for papers and, most recently, also for scientific data. “I feel like I can be a student all over again,” says Professor Idan Segev, a founding father of computational neuroscience and a board member of the OBI. “Turning electron-microscopic images of the brain into functional computational models is something even Ramón y Cajal, the founding father of neuroanatomy, could not have dreamed about,” says Professor Javier de Felipe, Founder of the Cajal Blue Brain initiative and board member of the OBI.

The OBI will host and support labs across the levels of brain organization. Sub-cellular labs will allow modeling of the brain at the molecular level with detailed models for the brain’s metabolic system and for ion channel models prescribed by genes, paving the way to replicate neurons at the genetic level. “For the first time, we can build circuit models that incorporate blood flow and metabolic energy constraints. We also show how ion channels encoded by the genes expressed in each neuron come together to form the electrical behavior of the neuron.” says Dan Keller, in charge of supporting the Sub-Cellular Labs.

Cellular labs will allow simulation neuroscience on single neurons with all their incoming synapses—from any region of the mouse brain, selected regions of the rat brain, and all types of neurons in the human neocortex. “We have turned the Harvard-Google electron-microscopic images of the human neocortex into functional neuron models,” says Marwan Abdellah, a legend in the Blue Brain team for capturing the exquisite morphological detail of neurons and their supporting cells, the glia.

The Circuit Lab allows users to call up and run any one of the many microcircuit models from various brain regions that were built by the Blue Brain, by the community, or from scratch. “Researchers can run an algorithm that computationally synthesizes, from a point in space, the full morphologies for any type of neuron and generate as many morphological variants as required to fill up a circuit with neurons,” explains Lida Kanari, who developed the breakthrough mathematical method and will continue supporting the Cellular Labs after joining Oxford University as a faculty member.

The Systems Lab allows users to select from a brain atlas as many brain regions as desired to form what we call a Brain System that can extend all the way to a whole brain. “We can now also synthesize not only the local axonal arborizations of neurons, but also the inter-regional axons connecting microcircuits to form a brain region—and even grow any neuron’s axons to span across the whole brain following major tracts connecting brain regions,” Kanari adds.

“Some physicists criticized that such models would never work like the real brain because of a potential explosion in the number of free parameters. But when the proper techniques are used, they are not free parameters, but biologically prescribed and with strong inter-dependencies. So, free parameters, so free parameters decrease exponentially the closer one follows biology. The Blue Brain has published many papers that show that model neurons and brain tissue constructed in this way closely mirror the behavior of real neurons and brain tissue, replicating what biologists observe in their experiments. And when it does not, it is not a failure—it is an exciting moment, a potential breakthrough because it means we are at the very frontier of our knowledge. It can point to what is missing or what is not understood properly, guiding the next breakthrough,” says Michael Reimann, Director of the Blue Brain Project’s Simulation Neuroscience division and Chief Science Officer of the OBI.

“We made a huge number of breakthroughs, published in several hundred papers, in this way. For example, this is how we found a key factor that switches the behavior of brain circuits from an awake-like state, into a sleep-like state with implications for our understanding of learning, we discovered how neuron connectivity shapes their collective behavior, how diversity of morphological and electrical properties makes the brain more resilient, and gained a deep insight into how a group of neurons processes information and more. When we recreated synaptic plasticity mechanisms in this way, we also found that it unified many theories and resolved many discrepancies in how synapses learn,” he added.

“It’s a very exciting time for neuroscience. I hope researchers leverage this opportunity and push the boundaries, both of the virtual lab capabilities and the OBI team, and challenge the community to deliver even more tissue-accurate digital brain models and more powerful software tools to accelerate everyone’s research—no matter which area of the brain or which species they are working on.” says Markram.

“The OBI is launching a freelance paid program, inviting computer scientists, physicists, and computational neuroscientists from around the world to join virtual labs and help unlock the full potential of simulation neuroscience. Additionally, OBI will offer commercial virtual labs for industries to prototype neuro-prosthetic devices, explore brain diseases, test treatments, and design targeted drugs. The possibilities are endless, and we encourage everyone to create a virtual lab and embark on this exciting journey with us.” concluded Georges Khazen, OBI’s CEO.