By Aaron J. Sender
What’s the difference between a protein and a robot? Not much, says Texas A&M computer scientist Nancy Amato. For the first few years of her career, Amato’s interest lay in planning paths for robots around obstacles to get them to their destination. But this work soon led her down a path she could never have planned: studying how proteins fold.
It turns out determining how to get a multi-jointed robot to, say, navigate through the narrow passages of a car engine and pluck out a sparkplug is remarkably similar to describing how a string of amino acids folds into its final, active three-dimensional form. In fact, says Amato, “We use the exact same code.”
To translate the motion planning programs, as roboticists refer to them, Amato and her graduate student Guang Song simply replace the restrictions set by a robot’s joints with the limits inherent in a protein’s atomic bonds. And instead of directing robots around physical obstacles, their program shuns prohibitively high-energy states.
But there are other differences they must deal with as well. “When we were working on the motion planning problems, usually we just cared whether or not a path existed. We didn’t care which path we found,” says Amato. As long as the robot could get to its destination, “I didn’t care if there were hundreds of paths — if my planner gave me back one, I’d be happy.” When applied to protein folding, however, she must choose the path a protein would actually take.
To do this Amato first generates what is called a probabilistic road map. The map contains sample points, representing random intermediate conformations, en route from the denatured protein to its active form. The computer then calculates all the possible routes and determines the best one. “If you are looking for the shortest path from New York to Texas you would add up the miles on all the different routes,” says Amato. “In this we do something analogous. But the weights we assign don’t relate to distance; they relate to energies.”
Just as interstates find the best way through the Appalachian Mountains, as a protein folds it prefers to circumvent high-energy peaks. To calculate the potential energy of the protein at various possible points along the way, Amato factors in hydrogen and disulfide bonding, van der Waals interactions, and hydrophobicity.
Currently Amato can determine the path of a 150-residue protein on a single PC in a few hours. Now she is working to knock that down to minutes or seconds by parallelizing the algorithms to run on many processors simultaneously.
In trying to understand how proteins fold, Amato is more concerned with the journey than with the final destination. “We’re not doing structure prediction,” she says. She believes that the insight gleaned from her work may eventually provide researchers with clues about how to fix proteins that misfold and cause disease. “If there’s something that gets stuck in some intermediate, you may be able to somehow alter it so that it doesn’t happen.”
Since she presented her technique several months ago at the RECOMB meeting in Montreal, she’s garnered lots of attention from pharma and biotech companies. “Every week I get a few inquiries,” she says. “But I’m more excited about having interested [academic researchers] in the biochemistry side of the field.”