Protein Structure Prediction: Getting it Right
Using Rosetta@Home, a program that runs on the personal computers of 150,000 volunteers worldwide, David Baker’s team predicted the structure of a 112-amino-acid protein from scratch.
When nature folds an amino acid sequence into a protein, it usually knows that just one conformation is the right one. But when a computer tries to do the same thing, it often predicts multiple possible shapes. Now, a team of scientists at the University of Washington, led by biochemistry professor David Baker, PhD, have made a significant advance toward predicting which of the multiple structures is correct. They also accurately predicted a small protein’s structure without relying on X-ray crystallography.
To predict a protein’s structure, researchers must find the arrangement of the individual amino acids that represents the lowest energy form. It’s kind of like gravity, notes Baker. “If you drop a ball on a hill, it rolls to the bottom of the hill.” For proteins, that spot represents the most settled overall shape, a compact blob of amino acids linked into helices and sheets. In the past, it was hard to figure out when a predicted structure truly reached its lowest possible energy, not just an intermediate step. “If you drop a ball on a bumpy landscape, it may get stuck in a [higher] valley,” Baker says.
For a number of years, the Baker team’s primary tool for predicting protein structures has been Rosetta@Home, a program that relies on a staggering amount of computing power. “We employ the computers of about 150,000 volunteers,” Baker says. Volunteers install Rosetta@Home on their computers. It runs like a screen saver while the computer is otherwise idle. The program calculates many possible structures for an amino-acid chain and sends promising structures to the researchers. A central computer then searches for the lowest-energy structure, in which the chain curls up most comfortably.
His team’s new research, published online in Nature on October 14, 2007, describes a major refinement to Rosetta@Home that searches for a way out of “energy valleys.” The team fine-tuned how Rosetta analyzes the toughest protein sections. If the program consistently predicts the same folded shape, the answer is probably correct. But when Rosetta churns out many different solutions, the program now recalculates those error-prone regions in search of the lowest possible energies—and more robust final shapes.
The refined method makes it easier to get useful data from traditional protein-structure experiments, in which researchers blast X-rays at protein crystals. Baker’s lab also used the method to predict an accurate structure for a small protein (112 amino acids) with no X-ray data, an achievement noted in a Nature commentary as “a real breakthrough.”
The new research is “a significant milestone in the development of methods to model protein structure from amino-acid sequence,” comments John Moult, D. Phil., a professor of computational biology and biophysics at the Center for Advanced Research in Biotechnology in Maryland.
Rosetta@Home's clan around the world savors the success. As volunteer Antony Magnus wrote in an online message board: “I crunch for Rosetta because I believe in this project whole-heartedly.” Volunteers interested in participating in Rosetta@home can sign up at boinc.bakerlab.org/rosetta.