Nature 435, 720-721 (9 June 2005) | doi:10.1038/435720a; Published online 8 June 2005

Blue Brain boots up to mixed response

Jim Giles


Swiss supercomputer faces brainy task of modelling neural circuits.

One of the boldest brain-modelling projects ever attempted is about to get under way in Switzerland. A team of neuroscientists plans to use a supercomputer to create a biologically realistic simulation of the neural circuits responsible for higher mental processes in humans and other mammals. But experts elsewhere are divided about its chances of success.

Most brain models focus either on large- or small-scale features. Some teams have connected huge networks of simple units to recreate brain functions such as vision, for example, whereas others have built detailed computer simulations of individual neurons. Now a team from the Swiss Federal Institute of Technology in Lausanne (EPFL) will use the IBM supercomputer Blue Gene/L in an attempt to combine both approaches.

"This is one of the most ambitious computational-neuroscience projects ever planned," says Alain Destexhe, a modeller at CNRS, France's main basic-research agency, in Gif-sur-Yvette. "Blue Gene is one of the most powerful computers ever made available to neuroscience."

The Blue Brain Project will simulate part of the neocortex, the intricately folded region on the outside of mammals' brains. The Swiss team will model a column of cells from the rat neocortex, the animal for which the most detailed data are available. Each column is just 2 millimetres high and half a millimetre in diameter, but contains some 10,000 cells connected by 5 kilometres of fibres. In both humans and rats, these columns form a basic circuitry that is repeated across the cortex and controls everything from vision and movement to planning.

"It's as if in evolution this system was cloned and duplicated in mammals — it makes up 80% of the human brain," says Henry Markram, who leads the project at the EPFL. "The more we look at the neocortex, the more we are in awe of it."

IBM has built several Blue Gene computers for use in research (see 'Virtual Big Bangs and digital mushroom clouds'). The EPFL's version, which is worth US$10 million and should be switched on next month, is a rack of four refrigerator-sized units, each containing more than 1,000 processors. It can process over 22 trillion operations per second. If running today, it would probably be the fourth most powerful machine in the world.

Markram will feed the simulation with information from his group's database on the physiological and electrical properties and shapes of cortical cells, which is regarded as the most complete database of this kind in the world. Each Blue Gene processor will model one or two individual cells, which will then be connected up as in the rat cortex. The team will spend about three years testing and refining the system by comparing the results of simulations with experiments using real tissue.

Blue Brain boots up to mixed response


Makes you think: neuroscientists aim to create a computer model of a column of cortical cells.

Such a grand project is bound to attract critics, acknowledges Markram. "Half the research community will say this is nonsense," he says. Sure enough, when Nature put the idea to brain modellers, they hailed Blue Brain as a great leap forward in terms of the realism of simulations, but some doubted whether Markram has enough data on the cortex to make his plan work.

"This is an ambitious project that is bound to fail," says Terry Sejnowski of the Salk Institute for Biological Studies in San Diego, California. "We are still far from understanding enough about the brain to build a detailed realistic model."

Neuroscientists say that too little is known about the structure of the network connecting cortical cells, for example. They add that a truly realistic model would have to incorporate molecular activity in the regions where neurons connect, a level of detail that is currently beyond the Blue Brain Project.

Markram agrees that more data are needed, especially from the bottom section of the six-layer cortical column, but says that work to remedy this will be done while the machine is being tested. Once the model is fully functional, researchers will be able to play with it, tweaking the numbers of different types of neuron, for example, or altering the levels of certain neurotransmitters, to see how such changes affect larger-scale brain activity.

The machine could also probe conditions in which cortical circuitry seems to malfunction. For example, some symptoms of autism can be recreated in rodents by giving thalidomide during pregnancy. The thalidomide causes changes in the cortex cells that could be mimicked on Blue Brain, says Markram.

In the long-term, Markram has a grander plan that will raise eyebrows among even his most supportive peers. Within ten years, he predicts, column models could be duplicated and connected to create simulations of the whole cortex and eventually the whole brain.

Fred Wolf, a computational neuroscientist at the Max Planck Institute for Dynamics and Self-Organisation in Göttingen, Germany, is enthusiastic about the idea. But he adds: "Don't expect to see whole-brain simulations any time soon."