FoldIt Team/University of Washington

The ultimate puzzle. When it came to predicting the structure of this protein, humans bested a supercomputer.

Video Game Helps Solve Protein Structures

John is a Science contributing correspondent.

People playing a simple video game can match, and even surpass, the efforts of a powerful supercomputer to solve a fiendishly difficult biological problem, according to the results of an unusual face-off. The game isn't Pac-Man or Doom, but one called FoldIt that pushes people to use their intuition to predict the three-dimensional (3D) structure of a protein.

When it comes to solving protein structures, scientists usually turn to x-ray crystallography, in which x-rays shining through a protein crystal reveal the location of atoms. But the technology is expensive and slow and doesn't work for all proteins. What scientists would love is a method for accurately predicting the structure of any protein, while knowing nothing more than the sequence of its amino acids. That's no small task, considering that even a moderately sized protein can theoretically fold into more possible shapes than there are particles in the universe.

To get around that problem, computer programs focus on which shapes require the least amount of energy—and thus which ones the protein is most likely to adopt. But these programs must rely on intense computing to make any headway. One of the most powerful, Rosetta@home, was created by David Baker, a molecular biologist at the University of Washington (UW), Seattle. The program distributes its calculations to thousands of home computers around the world, automatically sending the results back to Baker's lab. (It runs on the same "distributed computing" architecture as the SETI@home search for alien life.) The entire network is capable of nearly 100 trillion calculations per second, dwarfing most supercomputers.

Two years ago, Baker wondered whether humans might help Rosetta@home do better. Although the program is impressively good at solving the first 95% of the folding of a protein, putting the correct finishing touches on a molecule often stumps it. People complained to Baker by e-mail that it was frustrating to watch the program flail around on their computer screens when the necessary final tweaks were sometimes obvious to the human eye.

So Baker teamed up with UW Seattle computer scientist Zoran Popović to turn protein folding into FoldIt, a relatively simple video game where people grab, poke, and stretch a 3D model of a protein, seeking to score more points by minimizing the protein's total energy. To evaluate this "human computing" strategy, Baker recently challenged FoldIt players to perform the final folding of 10 proteins. The true structure of each had been solved with x-ray crystallography, but the results were not yet released.

It was a close battle: The predicted structures from FoldIt players were closer to reality than those from Rosetta@home for five of the 10 protein structures. They were also significantly more accurate, Baker's team reports in tomorrow's issue of Nature.

FoldIt has been a runaway hit, downloaded and played by over 100,000 people since it was released in May 2008. Baker and Popović are now studying the top players in the hopes of teaching computers the humans' tricks.

When the game debuted, Arthur Olson, a molecular biologist at the Scripps Research Institute in San Diego, California, told Science that he doubted that nonscientist players could get
very far. "I'm thrilled to be wrong," he now says. "What I didn't know is that this game would actually create experts." Olson expects that computers will eventually learn enough from the human players to beat them, as IBM's supercomputer Deep Blue did with chess. But because human spatial reasoning is still so much better, he says, "computers still have a long way to go."="#320>

Posted in Math