Among the topics that generated significant buzz at SLAS2013 was Foldit, a videogame that is being used to crowdsource potential solutions to complex biochemical problems. Foldit co-developer Seth Cooper, Ph.D., creative director at the Center for Game Science at the University of Washington, Seattle, gave a presentation on the game, which has garnered a number of awards, including an honorable mention for technical excellence at the Independent Games Festival 2013; a Katerva award 2012 behavioral change category winner, first place in the interactive games category at the NSF International Science & Engineering Visualization Challenge 2011 and Innovation of the Year, TechFlash Newsmaker Awards 2011.
The concept for Foldit started with David Baker, a professor in the university's biochemistry department; Zoran Popovi? and David Salesin, professors in the department of computer science and engineering, and Zoran's students, Seth Cooper and Adrien Treuille (now a professor at Carnegie Mellon University). "I've been playing games since I was a kid, so when we first started talking about the project, I thought it was a cool opportunity both to work on a game and also show that games can be more than just distraction or entertainment—they can do something useful," Cooper says.
Several projects endeavoring to involve the general public in solving science-related problems already existed, and still do. Among the best known are SETI@home, which is trying to uncover extraterrestrial communication, and Rosetta@home (developed by Baker), which is trying to uncover novel proteins that may have relevance for drug discovery. However, both of these initiatives use distributed computing, a collaborative effort in which individual personal computer owners allow some of their computer's processing time—generally "down" time, when the person isn't using the computer for other purposes—to be put to the service of problem solving. In this scenario, the individual essentially is a passive viewer of a screensaver that shows the computer trying to identify potential drug candidates.
By contrast, "instead of asking people to contribute their spare computing power, we wanted to give them a way to contribute their spare thinking or problem-solving power—their mental power that wasn't being used in other projects," Cooper explains. "People running Rosetta@home can watch the screensaver and probably see a number of things that won't work, but there's no way to intervene. So one of the challenges in designing a game like Foldit was figuring out what part of the problem solving computers can do and what part people can do, and how to put those two complementary pieces together in such a way that we can do something we wouldn't have been able to do with each of them separately."
The fact that they succeeded undoubtedly is a good thing, since there is no shortage of protein-folding problems to solve. The human body has more than 100,000 different kinds of proteins, and although the genetic sequences of many of them are known, how they fold up into complex shapes to assume various biological roles is much less clear because "even for relatively simple proteins, there are just too many possible shapes for a computer to search through," says Cooper.
"Because protein folding is both spatial and structural, it's all about fitting pieces together in just the right way," Cooper continues. "Therefore, a Foldit puzzle looks like a kind of 3-D Tetris, where the goal is to take all the pieces of protein and fit them together so that the resulting protein is very compact, with no empty space in the middle."
Not surprisingly, when Cooper and his colleagues launched the game in 2008, "many people were skeptical that something so engaging, that could be played by anyone, not just scientists, could actually work. They thought it was a cute idea, but that it wouldn't do anything real," Cooper observes. "But we've shown that it certainly can do something real." In 2010, a paper published in Nature provided a proof of principle that, according to the authors, "complex scientific problems can be solved with human-directed computing" and that the multiplayer games such as Foldit represent "a powerful new approach to solving computationally-limited scientific problems."
Subsequent published papers documented that Foldit players were able to solve the crystal structure of the Mason-Pfizer monkey virus retroviral protease, providing insights useful in the design of antiretroviral drugs; develop and share new protein-folding algorithms; and remodel the backbone of a computationally designed bimolecular Diels-Alderase to increase its activity and functionality.
In the PNAS paper, the authors reported that Foldit players developed more than 5,400 folding strategies, two of which continually rose to the top. As stated in the paper, "Examination of the algorithms encoded in these two recipes [folding strategies] revealed a striking similarity to an unpublished algorithm developed by scientists over the same period. Benchmark calculations show that the new algorithm independently discovered by scientists and by Foldit players outperforms previously published methods."
In the Nature paper, Cooper and his coauthors conclude: "The Foldit community's ability to successfully guide large-scale protein design problems demonstrates that biophysics expertise developed through game playing transfers effectively from structure prediction to more open-ended design challenges."
While the goal of Foldit is to contribute to important scientific research, the game itself has to be engaging enough to garner the large numbers of committed players needed to solve problems. A testament to that engagement is the fact that there have been over 300,000 players, working alone ("soloists") or in groups to solve various protein folding problems ("puzzles"). The Foldit multiplayer game site fosters both collaboration—groups, blogs and forums—and competitiveness, by showing rankings for top "evolvers" (people who have improved solutions provided by others) and soloists, as well as global points (accrued after a puzzle closes) and other measures of achievement.
Tutorial and introductory puzzles bring novices up to speed on the basic concepts and tools needed to engage in the science puzzles. When tackling science problems, users are urged to keep in mind three principles or "rules" that affect their score:
1. Pack the protein to avoid empty spaces where water molecules can enter.
2. Hide the hydrophobics—sidechains that should not be touching the water that surrounds most proteins; hydrophilic sidechains, by contrast, should be exposed as much as possible.
3. Clear the clashes that occur when two sidechains in a folded protein are too close together; intersecting sidechains lower a player's score.
Beyond these basic principles, each puzzle has its own parameters. "We work with biochemists who are looking at a particular protein structure, for example, and we take the problem they're working on and try to translate it into something that can be expressed in the game," Cooper explains. "We might say things like ‘these parts of the protein can move, these parts can't move at all and these parts can be cut apart and restructured.' We also provide tools players can use to manipulate the protein, as well as automated methods to optimize it."
Problems that are turned into puzzles generally originate with University of Washington scientists, although some come from other researchers—for example, the monkey virus protein, which both experimental biochemists (at Adam Mickiewicz University in Poland) and computational biochemists (at UW) had been working on for more than a decade. "The Foldit players tend to have new ways of looking at problems that may not have been considered before," Cooper says. "In this case, they solved it in three weeks, which is a bit longer than we usually run puzzles."
By contrast, when working on the Diels-Alderase enzyme puzzle, the players and the biochemists went back and forth, working together to refine the design of the enzyme. "It became an iterative process, where scientists would get input from the players and work on designing the enzyme until they wanted some new ideas or felt like they were stuck and needed some additional input," Cooper continues. "They tested some of the Foldit proteins in the lab to see how they worked, and provided information and analyses that we could feed back into the next round of that puzzle so the players could continue to work on it."
While working on puzzles, players—whether working alone or in a group—try to get the highest possible score. That score is based on an energy function developed by biochemists that roughly corresponds to how well folded the protein is. "The better folded the protein, the more stable it is, and the higher your score is," says Cooper. The Foldit site also has a leader board that displays the score of every person working on a particular protein at the time a player is online, so anyone can compare and see how well he or she compared with others.
Given Foldit's successes and the emphasis on scoring, one would think many of the players are working scientists. However, recent surveys suggest only about 25 percent have had any scientific training beyond a college class in biochemistry. "Foldit isn't about knowing a lot of chemistry. It's more about the ability of the players to look at the structures and see how the pieces fit together," Cooper observes. "The game taps into their intuitive 3-D spatial skills—their ability to rotate chains of amino acids in cyberspace. Beyond that, it's difficult to generalize about a ‘typical' player. A survey we did a while ago suggests they're all ages, from all over, and with different educational backgrounds and careers."
In addition to having spatial awareness skills, "you have to be open and willing to share what you're working on and credit for what you find," Cooper continues. When a problem is solved, players get credit individually for what they've contributed and the groups they worked in also get credit. What's more, that credit is shared in scientific publications. When the monkey virus protein puzzle was solved, the Foldit developers said they were going to write a scientific paper and asked the players who worked on that puzzle if they wanted to be listed as coauthors. Instead, they chose to be listed as a group. "Although game players don't necessarily want to be credited for findings in the same way most scientists do, we feel it's important to respect their contribution and credit them for their work," Cooper emphasizes.
Now that Foldit's capabilities have been demonstrated, Cooper and his colleagues are working on expanding its reach while also developing new audiences for the game. The current version of Foldit runs on PC, Mac and Linux operating systems. "We're now working on a version for the Kinect gaming platform, so people don't have to use a mouse and keyboard to interact with the protein; instead, they can actually grab the parts in three dimensions and move them around," Cooper says. Currently there are no plans to try to create a mobile version because the game is requires significant computer power to model and optimize the protein. For a mobile version to work, says Cooper, "we'd have to restructure Foldit, probably with a cloud-based kind of architecture, so the computation isn't done on a phone, but somewhere else that has the resources to do it."
The team also is looking at ways to use Foldit for educational purposes. "We didn't design the game with education in mind, but we did try to make it fun and accessible. That has had the side effect of making Foldit a great way to get introduced to protein-folding concepts," Cooper says. "We've been contacted by some teachers who are starting to use Foldit in the classroom, and by others who ask us how they can use it in the classroom. So we want to create a way to crowdsource ways to integrate Foldit into the curriculum so that teachers who are already using it can help others fit it into their lesson plans."
Meanwhile, the Center for Game Science is developing other games focused on different aspects of education, including fractions and other concepts introduced in early math classes. "With these kinds of games there are always two aspects—the engagement that motivates people to want to play the game and spend their time on it, as well as the aspect of doing something real. While this kind of open, crowdsourcing approach clearly won't solve every problem, for particular problems it's a really useful and powerful way to help scientists advance what they're working on," Cooper observes. "Similarly, it's a way to stimulate problem solving in the classroom, which then could carry over into the real world."
Any scientist, regardless of affiliation, who is interested in taking advantage of the problem-solving potential of Foldit is invited to contact Cooper directly or via the Foldit website. "We like to provide background on the problem and explain why it's important. Although we're focused mostly on biological structures, other games are now available to handle other kinds of problems, and we can refer people to those when applicable," Cooper concludes.
Readers also can see Sidelines with this article for direct links to other types of scientific games.
May 13, 2013