Unraveling the Mysteries of the Brain
When Michael J. Fox was 30 years old, he started to twitch.
In his 2002 memoir, Lucky Man, the actor recounts the moment that would set the tone for the rest of his life. Fox, most famous for his role as Marty McFly in the Back to the Future movie franchise, was in Florida shooting a movie. It was early morning and he was profoundly hung over. He flung his hand over his eyes to block out the sunlight filtering through the window of his hotel room.
And then he felt it.
Tap, tap went his pinkie against his cheek. Tap, tap, tap. A light but steady thumping so reminiscent of a moth’s wing that Fox prepared himself to smack the bug away.
But it wasn’t a moth.
Fox shook out his wrists, pulled on his fingers, and then finally resigned himself to the brief spasm that was overtaking his pinkie.
But it wasn’t brief. The twitch never went away.
It was early-onset Parkinson’s. A creeping disease, Parkinson’s is no less indomitable for taking it slow. By the time Fox’s pinkie started drumming its strange melody on that sunny Florida morning, his brain was almost entirely divorced from his will. The tremor was just the effect of a separation that had started long before and was almost complete, like the imprint of light you see in the sky after a firework explodes.
Moths. Divorce. Fireworks. These metaphors are all attempts to navigate the gray mystery of the brain.
For Dr. Xiaohui Chen (left), however, the brain isn’t any of these things. Instead, Chen, an Assistant Professor in Statistics at the University of Illinois at Urbana-Champaign, sees a matrix.
“I am working on high-dimensional statistics, in particular focusing on large dimensional matrix estimation. One project that I have been working on with my student, Xin Ding, is to estimate the time-varying transition matrix for high-dimensional vector autoregression (VAR) model for non-stationary time series data,” he says.
Chen is interested in discovering connections between large amounts data. And in generating columns and rows of data, Chen and his team are helping neuroscientists translate the mystery tapped out by Fox and one million other Parkinson’s patients in the U.S.
They’re looking for the mechanism that makes neurodegenerative diseases tick.
Picture this. A doctor rolls a patient into a large metal tube.
Snap, snap goes the Magnetic Resonance Imagining (MRI) machine, taking images of the patient’s brain. Snap, snap, snap. The strong magnetic field and radio waves that run the machine let it capture images of blood flow in the brain. Protons line up like soldiers, are zapped by radio waves, scatter, and rebuild, this time with signals that are recorded by the MRI. Where the blood is, there, too, are data.
“When doctors perform brain scans, they take pictures of activities in your brain and over many parts of it. That’s a lot of pixels, and each pixel is a variable. Repeat it many times and you get a matrix,” Chen explains.
But nothing is easy, and that goes double for brains. In particular, brains are noisy places.
“With biomedical data, the signal-to-noise ratio is low. The signal gets buried in a lot of noise,” Chen explains.
In Statistics, “noise” is the static that obscures the valuable data, which is called the “signal”. Noise is caused by a lot of things, namely, missing values (picture blank rows in a spreadsheet), outliers (weird data that behaves differently from the majority), and heat. Any large data set is going to have noise, and it has to be processed before a statistician can find the signal.
The challenge of tracking down the signal is one of the things Chen likes so much about his research.
Chen has a Ph.D. in Electrical and Computer Engineering, but is more comfortable calling himself a statistician. ECE relies heavily on statistics, and Chen found that he was personally drawn to projects involving high-dimensional statistics with complicated spatio-temporal dependent structures. Like brains.
“Each part of the brain is a spatial variable. Every variable is measured over time. This creates rows and columns that are cross-correlated,” Chen says.
For researchers like Chen, the brain is a series of regions strung together by high-voltage power lines. In these regions live millions of variables, all leaning over their fences to gab with each other in impossibly fast dialects.
To eavesdrop on their conversations, neuroscientists take images of different parts of the brain and at different times.
But how does one person—or even a team of neuroscientists—even begin to understand the conversations the brain holds with itself in the dark?
One person can’t. A team of brain doctors can’t.
How about one person, a team of medical collaborators, and a supercomputer? That will do it.
Deus Ex Machina
Chen uses the Illinois Campus Cluster Project (ICCP) to process functional Magnetic Resonance Imaging (fMRI) data given to him by medical collaborators.
The ICCP provides researchers at the University of Illinois with high-performance computing resources to store and process large data sets.
Chen uses the ICCP’s supercomputers to create a model of connectivity patterns in the brain, or what part of the brain is affecting other parts at any given moment. Chen is then able to provide a network of the brain to neuroscientists.
Keeping track of all of these connections happening all over the brain across time, needless to say, would take forever on a desktop.
“There are just too many variables,” Chen says. “We’re running simulations to learn how variables relate to each other. Without ICCP, these simulations would take forever. With ICCP, most simulations take between just three or four hours to a few days.”
This is great news for medical collaborators who are searching to find better medicines to treat Parkinson’s patients. They provide Chen with many images of patients’ brains both before and after treatment.
“They want to know if the medicine is working and how it’s working,” Chen says.
As a statistician, Chen is able to use ICCP to state when he sees a change in the data. When Chen gives the analyzed data back to doctors, they can use them to determine how best to treat their patients. And they’re a little closer to understanding how Parkinson’s works, which means they’re closer to understanding how to cure it.
With each model, we’re one small step closer to unraveling the greatest mystery of all: ourselves.
The Illinois Campus Cluster Program (ICCP) is the centralized hub of supercomputing resources at Illinois. Researchers from every field, as well as individuals, groups, and campus units, are welcome to invest in and use these resources. Researchers use the Campus Cluster for a variety of projects, including statistical modeling and data visualization. For more information, see https://campuscluster.illinois.edu/ or contact email@example.com.