#1037 2011 · Foldit (University of Washington Center for Game Science) · Biotechnology / computational biology
A protein stumped scientists a decade; turned into a game, strangers solved it in weeks
the problem
A retroviral protease structure, key to AIDS drug design, had defeated automated folding algorithms for over a decade
background
The structure of the M-PMV retroviral protease, an enzyme central to how a monkey AIDS-like virus matures and replicates, had eluded structural biologists for more than ten years. X-ray crystallography gave incomplete data, and the computer algorithms researchers used to infer 3D protein shapes from that data — exhaustive, brute-force searches through astronomical numbers of possible foldings — kept failing to converge on a plausible answer. Solving it mattered directly for AIDS drug design, since knowing the enzyme's shape reveals where a drug could bind and disable it.
The University of Washington's Center for Game Science had already built Foldit, an online game that let anyone manipulate a 3D protein model by hand, competing for score against other players and against the computer's own folding algorithm, Rosetta. Researchers who had run out of algorithmic options handed the unsolved M-PMV structure to Foldit's players as a puzzle, betting that human spatial intuition might succeed where automated search had not.
what everyone would do
The available options were to keep running more exhaustive computational searches with faster hardware or refined algorithms, or wait for better crystallography data — both continuations of the same automated approach that had already failed for over a decade.
what they saw
Researchers saw the failure wasn't a lack of computing power but a mismatch of tool to task: humans are unusually good at 3D spatial reasoning, the exact skill the folding puzzle needed most.
the move
University of Washington scientists loaded the unsolved M-PMV protease structure into Foldit as a competitive puzzle for the game's players, who used the game's manipulation tools to physically twist and fold the 3D model by hand, competing for the highest-scoring configuration against each other in real time.
why it works
Framing the folding problem as a scored, competitive puzzle converted an abstract scientific question into something with instant, legible feedback, which is what makes a game engaging enough for strangers to spend real effort on for free. Because Foldit already had an active player base practiced at manipulating protein shapes, the platform could apply thousands of parallel human search attempts to the same structure simultaneously, each informed by players' accumulated intuition from prior puzzles, succeeding where a single automated search process, however fast, kept getting stuck.
the payoff
Foldit's Contenders team solved the M-PMV protease structure in about three weeks; the result was published with players as co-authors.
where it breaks
The approach depends on the underlying problem being one where human spatial or pattern-based intuition genuinely outperforms algorithmic search, and on the problem being translatable into a game with clear, immediate scoring feedback; a problem with no visualizable structure, or where verifying a proposed solution is itself as hard as finding it, won't gain much from crowd play.
what came after
The result became a landmark case for citizen science in molecular biology, demonstrating that non-expert human pattern recognition could outperform pure computation on certain structural problems, and Foldit has since been used on other unsolved protein and enzyme-design challenges by the same research group.
references
- [1]Gamers succeed where scientists fail: Molecular structure of retrovirus enzyme solved, doors open to new AIDS drug designScienceDaily, 2011sciencedaily.com
- [2]Foldit Gamers Solve Riddle of HIV Enzyme within 3 WeeksScientific American, 2011scientificamerican.com