Supercomputing Project Aims to Simulate Human Brain

Supercomputing Project Aims to Simulate Human Brain
Stefan Lovgren
for National Geographic News
July 20, 2005

It may be the most complex structure in the universe, a tangled web of more than a hundred billion nerve cells. For centuries, scientists have studied it, yet very little is known about the way the human brain really works.
Now powerful new computing technology is enabling scientists to learn more about the brain than ever before.
At the forefront is a new initiative to create a software replica of the brain's neocortical column, the smallest network of neurons (nerve cells) and an elementary building block of the mammalian brain.
The project is seen as a first step toward the long-term goal of creating a 3-D computer simulation of the human brain.
"We are not trying to build a copy of the human brain, or some magical artificial intelligence device," said Henry Markram, who heads the Brain Mind Institute at the Ecole Polytechnique Fédérale de Lausanne in Switzerland. His laboratory is collaborating with the computer giant IBM on the project.
"This is really a discovering of how the brain works," Markram said.
10,000 Trees
More than a hundred billion neurons make up the human brain, and the nerve cells are bunched in neocortical columns. These columns mark a jump in the brain's evolution that occurred 200 million years ago as mammals emerged from reptiles.
Since then the columns have multiplied within the mammalian brain to make more powerful minds.
In primates, and especially humans, this replication continued at such a rapid pace that the neocortex, the largest and most complex part of the brain, folded in on itself to make space for new columns. This is what gives the human brain its wrinkled shape.
The discovery of the neocortical column—which is half a millimeter (about two-hundredths of an inch) in diameter and two millimeters (about eight-hundredths inch) long and contains about 60,000 neurons—earned Torsten Wiesel of Rockefeller University in New York the Nobel Prize in 1981.
While scientists are able to make computer simulations of individual neurons, they have not been able to mimic the neocortical column, simply because of its complexity.
"Think of a forest, then imagine taking 10,000 trees and squeezing them together until there is essentially no space between them. That's what the neocortical column looks like," Markram said.
Blue Brain
To untangle this neurological forest, Markram is extracting tissue from the brains of rats, then subjecting the neurons in the brain sample to electrical stimulation. This enables him to find out how the neurons are connected.
"To reconstruct this puzzle, we must know where exactly the neurons connect," he said. "My lab's effort is to work out the code of communication [between the neurons]."
The information drawn from thousands of these experiments will be plugged into a supercomputer, named Blue Brain, to make a software replica of a column of the neocortex.
The Blue Brain computer, which became operational two weeks ago, is the latest installation of IBM's Blue Gene supercomputing system. It has a peak speed of 22.8 teraflops, meaning it can make 22.8 trillion calculations every second.
"Until recently this type of project just wasn't feasible," said Charles Peck, who leads the Blue Brain effort at IBM's research division in Yorktown Heights, New York. "With the advent of Blue Gene, we now have the opportunity to extend the kind of science we do."
The computer model will then be checked in experiments against the neocortical columns taken from rats to see if the model matches the brain's biological behavior.
The template column was obtained from a two-week-old rat in the part of the neocortex that processes touch sensory information. The template will form the basis for building columns from other brain regions in developing and aging rats, as well as in other species, including humans.
Psychiatric Disorders
Once one column has been accurately simulated, the researchers will try to replicate several neocortical columns. By modeling other areas of the brain, they hope to eventually build a computer-based model of the entire brain. This will take at least a decade.
Some skeptics, however, doubt that it is possible to even simulate the neocortical column. Although Markram acknowledges the challenges facing his project, he says the huge leaps in computing power make the effort possible.
"We're sitting on a hundred years of data," he said. "If that's not worth anything, then we should be questioning what we've been doing for a hundred years."
Markram stresses that the project will not create artificial intelligence but aims to better understand and explore brain functions.
"It doesn't mean people should be afraid that their brains are going to be replaced by artificial brains," he said.
Instead, he says computer simulations of the brain may be used to study how and why certain microcircuits in the brain malfunction, which is thought to be the cause of psychiatric disorders such as autism, schizophrenia, and depression.
A brain research experiment that might take a full day, or even months, to conduct in a laboratory could potentially be done in seconds with a computer-based model of the brain.
"The ultimate goal [of this research] is to be able to ask questions about how the brain works that cannot be answered using traditional scientific techniques," Peck, the IBM scientist, added.