Proteins are the workhorses of biology. Among their many functions, proteins speed chemical reactions, enable blood cells to recognize intruding viruses, and copy DNA. The potential payoffs of making proteins that don’t exist in nature, such as those needed for HIV vaccines or as catalysts for more-efficient biofuel production, are huge. But making proteins to meet a specific need can be difficult.

Now a leading protein researcher has teamed up with computer scientists to create an online game for developing useful protein structures. David Baker, a leading protein scientist at the University of Washington, says that players will help his lab design new vaccines and make enzymes for repairing DNA in diseased tissues.

For years, biochemists have reengineered naturally occurring proteins by growing them in viruses and single-celled organisms in a process called directed evolution. But researchers need to start with a preexisting protein, which makes it difficult to develop proteins with totally new functions. In a major step forward, Baker recently demonstrated the first algorithm for building novel, functioning enzymes from scratch. But while proteins built from the ground up may have chemical properties unmatched by anything in nature, they aren’t particularly efficient.

The game, called Foldit, is part of Baker’s vision for the future of protein engineering. His algorithms are good at the nitty-gritty of generating completely novel protein sequences for a particular purpose. But humans, who are better at seeing the big picture than computers are, could improve computer-designed proteins by playing the game.

Proteins are made up of long strings of amino acids that are folded up into complex three-dimensional tangles with many subregions. The function of a protein is dependent on this three-dimensional structure. One pocket might be ideal for grabbing on to another protein, for example. Other parts of the protein may play a purely supportive, structural role, holding the molecule together. Baker’s new method for creating novel proteins begins with the active sites. Once they’re in place, structural concerns, especially how tightly packed the protein is, determine whether the design is feasible. Figuring out the best way to hold together the active sites is a complicated search problem that requires a lot of processing power. There are a myriad of possibilities, but most won’t work.

Since 2005, Baker’s lab has relied on the computing power of Internet users who’ve installed a program that searches through protein designs. About 200,000 people around the world run his Rosetta@home program when their computers would otherwise be idle. As Rosetta@home runs, it displays the protein structures that it’s processing as a screen saver. Some users staring at the structures, which look like large tangles of multicolored string, told Baker that they could see how to make the structure better–where to tuck in a loose end, or how to pull the structure together tighter–but were frustrated because they had no way to provide input.

When the computer doesn’t know what the best next step is, it changes the structure randomly. Baker says that he began to wonder whether people working with computers can solve a hard problem that computers can’t solve alone.

Baker enlisted Zoran Popović, a game designer at the University of Washington, to create the protein-folding game. This task presented Popović’s design team with an unusual challenge. “For standard games, the goal is known, and the game play is designed to guide the player toward it,” says Popović. Mario needs to save the princess, collecting as many coins as he can on the way. But in the case of the protein game, the end goal is unknown. “The ultimate protein configuration, and how best to get there, are not known,” Popović says of Foldit.

The first several levels of Foldit are designed to teach players what good proteins look like and how to manipulate them using the tools of the game. Players can rotate proteins in three dimensions, pull together their constituent threads (called side chains), bend the protein’s overall structure (called the backbone), and try to generate hydrogen bonds, a stabilizing influence. The game graphically represents real protein chemistry. For example, it’s good to pack proteins tightly, but not too tightly: electrical charges in different regions of the side chains will repel each other if they’re too close to each other. Such clashes are represented by spiky, red burrs that disappear as the side chains are pulled apart.

After improving the designs of a few test proteins, players can advance into competitive play, working in teams or alone. Baker and Popović have set players to work on proteins whose structures are known in order to refine the game and train a group of players. In time, players will be working on new HIV vaccines and Baker’s other projects.

Luis von Ahn, a computer scientist at Carnegie Mellon University, agrees that humans bring problem-solving skills to the protein-folding game that computers can’t match. “The computer does a brute-force search, where we may know the shortcut,” he says. “We live in a 3-D world; we know how to navigate space.” Von Ahn has designed games that get people to help label images for Google and digitize books. Computers are bad at some tasks that are trivial for humans, such as recognizing a dog in a photograph or reading a blurry word. However, manipulating complex three-dimensional structures is a much tougher problem, and von Ahn worries that Foldit might be too difficult to gain a huge following.

Popović and Baker concede that the game is difficult. The aim is to make it entertaining enough that users will continue playing and get their friends to join in. By making the game available to anyone over the Web, the researchers expect to find people they call protein savants–people who are very good at solving protein structures and who will spend several hours a week playing the game.

Popović says that the designers will continue to improve Foldit by logging and analyzing what good and mediocre players do. “Through analyzing how people play, we’re learning what the best players are doing and improving the game play with that information,” he says. Dane Wittrup, a chemical engineer at MIT who designs proteins, says that this is a promising approach. “I suspect that if they carefully analyze successful game-playing strategies, they are likely to learn new, automatable strategies that they’ll incorporate into their structure prediction programs.”

Starting this afternoon, anyone can download the Foldit game for free.