Bobby Kasthuri has a problem.
In an effort to understand, on the finest level, what makes us human, he’s set out to create a complete map of the human brain: to chart where every neuron connects to every other neuron. The problem is, the brain has more connections than the Milky Way has stars. Just one millionth of the organ contains more information than all the written works in the Library of Congress. A map of the brain would represent the single largest dataset ever collected about anything in the history of the world.
Making that map seems like a task that could consume not just one lifetime, but dozens. Yet in just three years, it might just be possible.
Kasthuri, a neuroscientist at Argonne National Laboratory, is one of many scientists whose research will use a new supercomputer the lab is building, which is scheduled to be deployed by 2021. The computer, called Aurora 21, will run one quintillion operations in parallel—a billion billion calculations—putting it on par with the processing power of the human brain. For the U.S., which has lagged behind China in an intensifying supercomputing race since 2013, this milestone—exascale computing power—is both a national status symbol and a scientific game-changer.
The demands for a simulation of the brain are immense, and just building a computer like Aurora 21 is a massive undertaking. The finished computer is expected to cost hundreds of millions of dollars. It will occupy around a quarter-acre, have thousands of miles of wiring, and, if supercomputer trends continue, draw as much electricity as a medium-sized city.
Aurora 21 is designed for more than just simulating our brains. It will be able to perform computationally demanding simulations for tasks as diverse as predicting the weather, tracing the evolution of the cosmos, and understanding how new medicines will interact with the human body.
Every computing milestone brings new possibilities for research, says Rick Stevens, Argonne’s associate laboratory director for computing, environment, and life sciences. But this one holds particular promise for neuroscience. In providing the capacity to simulate the brain, the supercomputer could illuminate the largely mysterious processes that underpin human learning, behavior, and even psychiatric disorders.
The promise of breakthroughs like these drive this arms race, says Michela Taufer, professor of high-performance computing at the University of Tennessee, Knoxville. The victor won’t necessarily have the biggest computer or the most medals, she says. “The metric of success is what kind of science you enable.”
Mapping the Brain
Early in his career, Kasthuri was struck by the fact that a fruit fly hatches from its egg fully competent, already knowing how to fly, while a human baby is born so utterly helpless that it’ll die of starvation without someone to feed it. He knew that this helplessness, the years we spend developing into functional adults, had to be key to what makes us human.
“We trade off being competent for being able to learn almost anything,” Kasthuri says. We’re born with a blank slate of some 100 billion neurons that get arranged and rearranged over time to create the hardware we run on.
But no one really understands how, exactly, that hardware is wired. So it’s at the level of these neurons that Kasthuri has set out to explore the brain.
With the help of Aurora 21, he’ll be able to piece together millions of two-dimensional images, reconstructing the brain in three dimensions—essentially creating a map, known as a connectome. In many ways, it’s like a city map. “That map of the streets in the city is going to tell you something about what that dynamic city is like,” Kasthuri says, like which parts sustain the most traffic and which parts are directly connected to others.
An exascale computer would be the first machine capable of crunching through such a massive amount of data at an efficient pace—in theory, letting scientists like Kasthuri map multiple brains. “There’s no way we want to do just one brain,” Kasthuri says. The most interesting findings, he expects, will come from comparisons. How does the connectome vary between two adults with different skills? Between an adult and a baby?
Kasthuri even hopes to compare a human brain with that of an octopus. Our last common ancestor was probably some worm-like creature that lived 600 million years ago, meaning that the octopus, which can learn and solve problems much like humans, evolved independently. But Kasthuri wonders what structural principles our brains share and what those principles reveal about how we think and learn. “Is there only one plan for a brain that can problem-solve?” Kasthuri wonders. “Or is there more than one way to skin that cat?”
That’s not something an unaided human could discern. It would be like looking at New York City’s roads, subway lines, air traffic, and shipping lanes and trying to understand where everyone is going. Fortunately, it’s just the job for a high-performance computer. Stevens, who has been working on plans for Aurora 21 since 2007, jokes that it helps that supercomputers don’t get bored poring over millions of images. “We need this kind of idiot-savant brain to understand the real brain,” he says.
Beyond the Structural Map
The structural map is just one part of the story, though. With just a street map, Kasthuri says, you never really know where traffic might build up or why. Likewise, the structure of the brain is just a starting point.
Kasthuri hopes to combine a structural map with collaborators’ maps of the brain’s electrical activity, or “traffic,” to see how the two together influence a person’s learning and behavior.
If successful, the technology could make waves in the medical field. Susie Huang, a radiologist at Massachusetts General Hospital and researcher with the Human Connectome Project, says that many disorders, such as autism and schizophrenia, are likely rooted in anomalies in the structure of neurons.
“A lot of how we diagnose disease is looking under a microscope and saying, ‘OK, these cells are altered, so therefore you have this kind of disease,’ ” she says. But MRIs and other current brain-imaging methods are too coarse and can’t easily suss out such anomalies. They can’t tie together cause and effect.
A fine-grained map of the brain could change that, she says, and help doctors diagnose psychiatric disorders or possibly even predict them.
A Changing Field
For other neuroscientists, like Columbia University’s Rafael Yuste, the most exciting part is not the map itself but how a national lab for neuroscience could transform the field. “Neuroscience has operated always a little bit like a ma-and-pa store,” he says, with small labs working within the limits of their budgets and the tools they can develop. But more recently, it’s begun to outgrow that model.
Kasthuri says that neuroscience has quietly evolved into a big-data field—“and we didn’t realize it as a community till five or ten years ago.”
Other fields have had to cope with similar growing pains. It’s a phase that the field of physics outgrew decades ago as researchers around the world started getting their data from large observatories and particle accelerators. Now, Yuste and Kasthuri believe, neuroscience needs to scale up, too.
Aurora 21 will help catalyze that transformation. It’s happened in other fields like physics, where massive, expensive tools push scientists to work together more by sharing time on the machines to gain access to their potential for discovery. In the process, those collaborations advance the field in a way that a lone machine or hundreds of independent scientists never could. Yuste hopes that this is the beginning of more collaborative and ambitious neuroscience.
Yuste led the team that first proposed a detailed map of the brain’s activity in the summer of 2011 at a meeting discussing the future of the field. He argued that the holy grail of neuroscience was to break the neural code—that is, to read the activity of every neuron in the brain and map that activity to a behavior. It was a goal separate from Kasthuri’s connectome but similar in scope. “Many people were very critical,” Yuste recalls. “They said that you couldn’t do it—it was impossible.”
Then, he says, George Church, one of the pioneers of the Human Genome Project, stood up from the seat next to him. Church said he’d heard these criticisms before. People had said similar things about the Human Genome Project, and they’d been wrong. Yuste says that’s when the conversation shifted.
Church is more modest about his role. “I’m not one to twist arms,” Church says. “There are a lot of things that I’ve run up the flagpole that, if nobody salutes, I just slink away with my tail between my legs.” But he says the moment was right for this one. The technology, the excitement, the ambition—they were all there.
Yuste and Church supported what ended up being the multi-billion-dollar BRAIN Initiative, an Obama-era grand challenge which funds research that attempts to understand how the brain works. Both Yuste’s and Kasthuri’s work on mapping the brain are just two of the “impossible” projects that the initiative has set in motion.
“It’s not exactly analogous, but I often think of the moonshot,” Kasthuri says. He thinks about how the average age of a NASA scientist was only 28 when the first crew landed on the moon in 1969 and about how the challenge fascinated a generation of scientists.
Kasthuri can’t be sure how his project will play out. In some respect, it’ll probably fail, he says with a laugh. “It seems enormous and monumental. A lot of those things don’t work,” he adds. “That’s just the nature of trying to do something incredibly hard.” But he’s inspired by having a challenge that captures his imagination.
For Taufer, the high-performance computing scientist, supercomputers like Aurora 21 swing open the door to possibilities that don’t exist in real life—the ability to test medicine, infrastructure, even weaponry free of the cost, safety, and ethical concerns that constrain real-life experiments.
But as grandiose as the possibilities are, Taufer emphasizes that the applications will work their way into our everyday lives, from predicting the weather to assessing the safety of our aging bridges and fighting common diseases like Alzheimer’s.
Despite the stiff competition between the U.S. and China, ultimately the supercomputer age is not about any one country or any one project, Taufer says, but about the diversity of science being made possible. “I look at this machine and I see a key to a solution.”