Foldit turns protein folding into a puzzle game — 57,000 players beat the computers

The University of Washington's Rosetta@Home screen-saver used volunteers' idle computers to fold proteins, but players kept emailing that the software was making obvious mistakes; the lab wanted to harness human spatial reasoning where computers fail.

Turn protein structure prediction into a competitive 3-D puzzle anyone can play — human spatial intuition, scored by biochemistry, feeds real research.

In 2008 Baker's lab and computer scientists Seth Cooper and Zoran Popović released Foldit: players twist and tug 3-D protein models with game tools, scored by Rosetta's energy function. In 2010 an analysis of 57,000 players showed humans beat the algorithm on 5 of 10 puzzles; in 2011 players produced an accurate model of M-PMV retroviral protease in ten days after a decade of failed attempts.

The 2010 Nature study found 57,000 players outperformed Rosetta on half the prediction puzzles; in 2011 players' ten-day model of M-PMV retroviral protease yielded the enzyme's crystal structure, giving researchers new targets for antiretroviral drugs.

In 2008 the University of Washington launched Foldit, a free online game that turns protein structure prediction into a 3-D puzzle. Players manipulate spaghetti-like protein models with a Tetris-like toolset, and every move is scored by Rosetta, the same energy function researchers use, so success in the game equals success in science.

The idea grew out of Rosetta@Home, a distributed-computing screen-saver that used volunteers' idle PCs to fold proteins. Its players kept writing in that they could see the software doing the wrong thing, so biochemist David Baker's lab teamed with computer scientists Seth Cooper and Zoran Popović to build a game around human spatial reasoning.

The approach worked: a 2010 Nature analysis of 57,000 players found people beat the algorithm on five of ten prediction puzzles and matched it on three more. In 2011, after scientists had failed for over a decade to solve the structure of M-PMV retroviral protease by molecular replacement, Foldit players produced an accurate model in ten days, leading to the enzyme's crystal structure and new leads for antiretroviral drug design.

It reframed a hard scientific problem as a competitive game, so thousands of untrained people engaged with it instead of one expert.

The scoring function was real biochemistry, not game fiction — high scores were better protein models, so play itself produced research data.

It exploited human spatial intuition, the exact skill pattern-matching algorithms lacked, as the 2010 head-to-head with Rosetta showed.

The 2011 protease solve proved the idea could beat professional methods: ten days for what molecular replacement had failed to do for a decade.

Make the scoring function the science and give amateurs real puzzles: thousands of untrained players' pattern-recognition skills become a research instrument, and participation is the contribution.

Foldit stayed online, and its players went on to help with proteins linked to cancer and Alzheimer's research, according to the University of Washington team. The 2011 M-PMV structure, published in Nature Structural & Molecular Biology, became the canonical proof that game players could solve real structural biology problems, and the game remains a touchstone for crowdsourced science — the moment play produced peer-reviewed results.