Behind the Scenes of Foldit, Pioneering Science Gamification
Proteins are involved in how our bodies develop and fight disease as well as how we behave and how we sense the world around us. Genes are instructions for making proteins, each of which is made up of some combination of building-block molecules called amino acids. Proteins can be hundreds of amino acids long, so they are complex and difficult to study. Although protein sequencing to find out the order of the amino acids in a particular protein is pretty easy for a biochemist, it is not simple for a biochemist to figure out all the possible shapes those amino acids can fold into. Foldit is a popular online citizen-science game, in which players are scored on the structure of proteins that they’ve folded. In Foldit puzzles, for example, players are rewarded for solving clashes and voids, places where the protein is not consistent with known biochemical patterns.
Seth Cooper is the lead designer of Foldit, and one of the original creators of the game. He is currently the creative director for the Center for Game Science at the University of Washington. In a recent Science Observer, American Scientist associate editor Katie Burke discussed Foldit and other citizen science games. The following is an extended version of her conversation with Cooper:
Compare playing Foldit to playing other games. What did you use as your inspiration, and what can a person who hasn’t played Foldit expect?
It’s most like a puzzle game. There’s not a lot of fast-paced action. There are no time-critical kinds of things, other than approaching the deadline for a particular puzzle closing. When we started, we often thought about it as a 3D Tetris. In Tetris you’re trying to fit all the blocks together to fill in all the empty space and remove the lines, and protein folding is a lot like that, except it’s in three dimensions. You want to remove all the empty space from the interior of the protein by packing everything in as tightly as possible. But there aren’t a lot of nonscientific games out there that are very similar to it that I know of. It definitely requires some kind of 3D spatial reasoning to follow.
You can also build your own protein, which could be a little bit like MineCraft, because there’s more creativity in the new design puzzles. You’re not just trying to solve a problem that already exists—it’s much more open-ended. You can build the protein up, more or less from scratch, starting with a scaffold that you can change to be whatever you want, building it with whatever little molecular subcomponents that you feel like. So there’s a lot of freedom and creativity.
What was it like to work on gamifying protein folding?
When we started out, we didn’t know what parts of the protein folding problem people would be good at, which parts people would find interesting, how much we would need to teach people about protein folding to allow them to be effective players, how we should make the protein look, what kinds of ways people should be able to interact with the protein. There were all these different variables and possibilities to consider. And it’s a double challenge when you’re making a game that has some kind of purpose, because you not only have to make a fun game, which just by itself is a pretty big challenge, but you’re also constrained by having to make a fun game that has some real scientific application. Before we released the game, we spent about a year coming up with prototypes. The game development team was going back and forth with the biochemists in the Baker Lab here at the University of Washington, seeing what they thought would be useful and watching people play the game to see where they got stuck, where they got confused, what kinds of things they liked to do and what kinds of problems they could solve. In a sense, we haven’t really stopped doing that.
We released Foldit about four years ago online, after a year of alpha testing. Instead of just a few people in the lab talking to people directly, we’re now using telemetry and analytics. The game records the structure and the moves that the players do, and we get data that we use to improve the game in every aspect, from the quality of the scientific results that are coming back to how long people play the introductory levels that are supposed to teach the game. The whole game is like an ongoing, continuous experiment. We release updates every couple of weeks with new features. Every week, we publish a different set of puzzles. So we’re adapting the game to the players to make it the most effective scientific problem-solving tool that we can.
What sort of people like playing this game, which requires quite a time investment to get to levels where they are working on a real-life protein?
A lot of different people are interested in playing the game. I think it’s much broader than what you might typically consider the core gamer demographic. There’s a pretty big international community. Most of the players are from the United States, but there are a large number from Europe, Australia, New Zealand and other countries. I think that a lot of the players are brought into the game or at least stick around for a number of reasons: They are motivated by the sense of purpose of contributing to science. It’s a game, but you’re not just playing a game. Something can come out of it, and we’ve shown that scientific results actually do come out of the game-play. A lot of the players, top players even, don’t have much background in biochemistry, and they’re still able to do well and solve interesting problems.
Do you get such a broad group because the game provides a variety of motivations, which in turn draw in a lot of different people?
We built the game using many different rewards or motivations. It's designed to encourage competition, because everyone is trying to fold their version of the protein better than everyone else to get the highest score. But it’s not just every person competing against every other person. There’s a lot of social interaction that’s supported by the game—chat and forums and things like that. Players can form teams to work together. So individual players can fold the protein for a little while, and then they can share that with other members of their group, who can pick up where they left off. The whole group gets credit for what the members have done, and the groups are competing against each other as well.
The leaderboard system on the website is also meant to motivate people. We support different skill sets, rewarding and recognizing players for doing what they’re good at. There are leaderboards overall for everyone, but there are also leaderboards for different types of puzzles, for individuals working alone and for groups working collaboratively.
There are puzzles that are just protein structure prediction, where you’re trying to figure out the shape of a naturally occurring protein. And then there are puzzles where you’re trying to design an entirely new protein. There are extra tools in the game that allow you to change the amino acids and the atomic structure of the protein, rather than just fold up some existing one. And we actually have a fairly recent type of puzzle, called “Symmetry,” in which there are multiple subpieces of the protein that are all exactly the same; they’re symmetrical. You can only change one of those subpieces, and when you change that one, all the others change in that same way. So we reward all the different kinds of skills that people can bring to developing solutions to each type of game. The game also looks fun and approachable to make players feel more motivated than they might when opening up a science textbook and seeing a standard picture of a protein.
What is the outcome when you design a protein? Is it synthesized?
In Foldit’s design puzzles, the players are able to build a hypothetical protein and see how it works in the game. The game’s score is based on a proxy for how well the protein would work in the lab, whether that’s how well it catalyzes some reaction that the scientists are interested in, or how well this protein sticks to some part of a virus, or even in the case of the Symmetry puzzles, how well the protein sticks to itself. Then we take the solutions that players come up with and present those to scientists for analysis. Solutions that are promising are then synthesized in the lab. We’ve been doing this for a fair number of the puzzles that the players have completed now.
We published a paper several months ago about a structure that the scientists and the players codesigned. The scientists were interested in an enzyme that catalyzes the Diels-Alder reaction, which brings together two small molecules to form a particular kind of bond that the scientists were interested in making. This catalysis would be useful for building other kinds of small molecules, such as drugs and chemicals. We went back and forth between the scientists and the players with several rounds of puzzles. In the end, we designed an enzyme that was about 20 times more efficient in catalyzing the reaction than the one the scientists had started with. And the really cool part of the solution was that the players had inserted about 13 amino acids, a really big departure from the structure that we started with. When the scientists looked at it, they said it was such a drastic change that it was something they wouldn’t possibly have considered, because they wouldn’t have thought it would work at all. But the players didn’t know that, so they just tried something that they thought would work, and it turned out to work really well. So they ended up with an enzyme that was much more efficient than, and quite structurally different from, the initial enzyme.
What are some other major breakthroughs that should be highlighted?
One really exciting recent result was with something in the game called “The Cookbook,” which allows the players to script or code up their strategies in the game. You basically write up what we call a recipe, which is an automated tool that will run moves in Foldit that the players have written. Players can share and modify these recipes online. We had this tool in the game so that we could learn strategies from the players and then automate those strategies. When we looked at the data, we found that one recipe was beating pretty much the next couple recipes combined in terms of popularity. This recipe was called “Blue Fuse version 1.1.” It was pretty simple, only a few lines. When one of the biochemists we were working with looked at it, he said it looked pretty familiar to him because it was similar to an algorithm that they were developing in the biochemistry lab, but they hadn’t published it yet. Organically, the community of Foldit players had come up with the same algorithmic moves that the scientist in the biochemistry lab had come up with independently. So we wrote a joint paper on both of the algorithms.
Have educators used Foldit? How does Foldit help players understand proteins?
We’ve been contacted by a fair number of teachers who are using Foldit for their classes. We know that sometimes they assign playing a puzzle for homework, so the students can see what proteins are like and get a sense of what they are before they learn the scientific details. We designed the game not necessarily to teach biochemistry but to teach the mechanics of how you would go about folding a protein. The game is a good starting place to encourage excitement about science and proteins, before getting into details of how proteins work. But one of the things that we’re starting to work on now is a version of the game that’s tailored specifically toward use in the classroom. We're supporting as much of the scientific curriculum as we can through interaction with proteins and other biological molecules. We’d like this version to be an effective teaching tool, while also setting up a way for educators interested in using the game to connect more easily. That way, when someone comes up with a good way to integrate the game into the classroom, other teachers could learn from that.
Do you think that humans will always be better than computers at folding proteins?
Once the players have the ability to automate some of the things that they do, some of the game becomes strategizing when you would run particular algorithms. If the players came up with an algorithm that solves protein folding, that would be really amazing. Proteins are very complex. It would take a long time, I think, for computers to catch up with people in many of these spatial reasoning and visual processing problems, because there are just so many variables. But even beyond that, players can be creative and come up with new things, which is even more challenging for computers to do. I think that thinking in nonintuitive and novel ways puts people even that much further beyond what computers are able to do.
What are the major changes that you’ve seen as the game has developed?
We started the game in 2007 and released it in 2008. The first Nature paper that we published compared the players’ protein folding with the biochemists’ state-of-the-art protein-folding algorithm. The game teaches some high-level rules, and one of them is that hydrophobic regions of the protein should be on the interior of the protein and away from water. In the game, hydrophobic and hydrophilic regions are color-coded. The players fixed incorrectly exposed hydrophobic areas better than the computational algorithm could. Soon after that paper, we discovered the shape of a retroviral protease enzyme that is associated with AIDS in monkeys. Then we added support for the recipes and the scripts, and we had the “Blue Fuse” result. We’ve added support for protein design, which resulted in the design of the Diels-Alder enzyme. Recently, we added support for the different leaderboards for the different puzzle categories. Since then, we’ve added a lot of new features, such as the “Symmetry” puzzles. We also added puzzles based on electron density, where, from some experiments, you can get a little information about where the mass of the protein is, and you can visualize that as a cloud that you want to fit the protein into. We just released a version of the game that supports Kinect. Rather than just using the mouse and keyboard, you can use Kinect to grab parts of the protein and push pieces with your hands. We wanted to try to support more natural 3D interactions and movements with the proteins.
What can we expect from Foldit in the future?
addition to novel protein designs, new symmetry games, the latest
interactions using Kinect, and innovative educational interfaces, we’re
also looking into new kinds of games beyond Foldit. One idea is based on
DNA and DNA nanomachines. You can use DNA as a building material and,
based on the preferences of the nucleotide-base pairings, you can cause
DNA to self-assemble into little shapes, machines and devices that move
and interact with other molecules. We want to build something inspired
by Foldit through which players build little pieces of DNA from scratch
or from some initial scaffold. These DNA pieces can be designed to do
things such as move around in the body, help to actively build up
structures or fight a disease. We want it to be really open-ended, so
that players can come up with things that we haven’t even thought of.
There’s really no telling what people will be able to do if you give
them the tools and the power to do it.