What if the brainpower used playing video games could be channeled toward something more productive, such as helping scientists solve complex biological problems?
A team of biochemists and computer scientists from the University of Washington (U.W.) in Seattle now reports that they have successfully tapped into this human problem-solving potential. Their competitive online game "Foldit," released in 2008, enlists the help of online puzzle-solvers to help crack one of science's most intractable mysteries—how proteins fold into their complex three-dimensional forms. The "puzzles" gamers solve are 3-D representations of partially folded proteins, which players manipulate and reshape to achieve the greatest number of points. The scores are based on biochemical measures of how well the players' final structure matches the way the protein appears in nature.
When the researchers analyzed the strategies employed by a group of 57,000 Foldit players, they found that humans were better at some aspects of pattern recognition and protein structure prediction than current computational software. In fact, the gamers outperformed the computer on five out of 10 puzzles and delivered similar results on three other puzzles by using more varied approaches to solutions not used by the computer. The findings appear online August 4 in Nature (Scientific American is a part of Nature Publishing Group).
The scientists hope to incorporate the newly identified strategies into computer algorithms for improved automated determinations of protein structure. The ultimate hope is to use these techniques to design new proteins to fight diseases such as Alzheimer's and cancer as well as develop vaccines against HIV and malaria.
"This is a new and exciting approach for tackling [complex] scientific problems," says Seth Cooper, a PhD student in computer science and engineering at U.W. and lead author of the paper. "We're getting people involved, people who don't necessarily have any training in biochemistry and science. Here's a way for anyone who has a passion and interest in science to help out just by playing a video game."
Proteins are the workhorses of the body—various types of proteins are involved in nearly every cellular process, from copying genes to digesting food. They move things around in the body and serve a major structural function—comprising a major fraction of muscle and skin, for example.
In order to carry out their myriad functions proteins interact with other molecules, and these interactions are highly dependent on a protein's 3-D shape. Improperly folded proteins are at the root of many illnesses, including some cancers and cystic fibrosis as well as Alzheimer's, Parkinson's and Creutzfeldt-Jakob diseases. Therefore, to better understand what a protein does, scientists have to glean its structure.
Understanding how proteins achieve their optimal, functional 3-D form is no simple task. In fact, in 2005 Science listed the "protein folding problem" among the 125 most compelling scientific problems. We have known for decades that the spaghettilike string of amino acids that make up a protein determines its 3-D shape. And thanks to the human genome project, we know all of the amino acid sequences of all the proteins in the body. There are, however, an astronomical number of ways in which any given string can fold into a 3-D structure, and only one of them is functional. How proteins in nature solve this problem and fold properly in a matter of mere seconds has puzzled scientists for decades. Were a protein to test out each of the possible structures one by one, it would take up to 1080 seconds for it to arrive at the correct structure, a length of time 60 orders of magnitude greater than the age of the universe. Clearly, there are some "rules" that govern how—and how quickly—proteins fold.
Scientists want to learn these rules and eventually be able to predict the 3-D shape that an unknown protein will adopt, information that could be used to develop new drugs to treat diseases or even to design new proteins with new enzymatic capabilities, such as synthesizing biofuels or degrading toxic substances.
One way to get this information is to use computers—testing out possible protein structures is just the sort of task that processors can handle efficiently, although it appropriates vast amounts of resources and still takes many years. To speed things up, scientists have developed protein-structure prediction software that relies on "distributed computing," a brute force approach in which many personal computers running the program donate idle CPU time to calculating protein-folding problems. In essence, a multitude of computers are banded together to form "supercomputer" that samples a great number of different protein structures.
Foldit emerged out of one of these distributed computing projects, Rosetta@Home, a screen saver that tries to find native protein structures by making random changes to the protein as quickly as it can, using as many personal computers as are available. Rosetta@Home was developed by a team led by David Baker, a co-author of the study and professor of biochemistry at U.W. "People started writing in saying that they saw the computer was doing the wrong thing, and wanted to know whether there was anything they could do," Baker says. His group suspected that they could improve the software's protein-structure prediction capabilities if they could add human reasoning and spatial problem-solving abilities, so they enlisted the help of computer scientists, including Cooper and his advisor, Zoran Popovi.
"In the distributed computational model a bunch of computers are running the same algorithm," Cooper says. "In the case of Foldit the game is distributed across a bunch of different players, and each of them has a unique perspective—their own algorithm, in a sense."
In their paper the scientists describe a particular class of partially folded proteins with a folding problem that the gamers appear to be particularly adept at solving. These proteins have both an exposed hydrophobic (water averse) area that "wants" to be buried in an interior region, and a convenient "hole" in the protein core in which to place the water-fearing domain. Cooper explains that this is a rather difficult problem for a computer to solve, and requires restructuring the protein in a way that is initially not favorable. "It's the kind of thing where a person can look at it and see [the solution]," Cooper says. Players were able to rearrange the protein to the more optimal shape, even though it required moves that caused players to lose points.
"They've successfully gotten a big part of man–machine communication working," says Peter Wolynes, professor of chemistry, biochemistry and physics at the University of California, San Diego who also studies protein-structure prediction but wasn't involved in the study. He notes that computers still perform better at coming up with possible structures from a featureless, unfolded amino acid chain. After computers render partially folded structures, however, this game "gives people an automated way of pulling [the proteins] apart. Once [players] know what the rules are and can visualize them, they can bring their intelligence to it," he says, adding that "part of science is recognizing parts that look bad," a task that is difficult to put into mathematical algorithms.
So what's next for Foldit? Cooper says they have given players the ability to design new proteins that do not exist in nature, and have begun testing the player-designed proteins in the lab. "Hopefully we'll end up with a player that will come up with cure for disease," Cooper says. "We haven't yet cured any diseases, but that's one of the goals. This is a chance to use video games for humanity, for a good purpose."