There are things artificial intelligence just can’t do, and protein folding is one of them. That’s why Fold.it and other resources are drawing on the puzzle-solving skills of a network of gamers to solve complex problems in protein folding. It’s an approach that is finally yielding its first concrete results.
To understand the nature of this challenge you need to understand a bit of biology. Proteins perform the vital tasks necessary for any living object, be it a human being or a plant. They carry oxygen, transport nutrients, power muscles, and speed up chemical reactions. They do all the work that keep organisms alive and functioning.
Each protein is a chain of linked amino acids. There are twenty different kinds of amino acid, with each defined by its particular combination and arrangement of carbon, hydrogen, sulfur, nitrogen, and oxygen atoms. These atoms come together in an unbranching chain to create a protein. Each main chain functions like a backbone, while each amino acid also has some dangling atoms, which form side chains.
These chains don’t like to just hang about in a single line. They tend to compress into a blob through a process of “folding.” The aminos for each different type of protein fold into the exact same configuration every time. Some amino acids are kept on the inside of the folded protein, while others remain on the outside. Some amino acids need to be near a particular type of amino, while others need to be kept apart. Each protein assumes the most efficient folded shape possible. When proteins fail to form correctly they can’t do their job, resulting in diseases like cancer, Alzheimer’s, and HIV/AIDS.
Proteins can assume incredibly complex structures. A simple one may have a chain of a hundred amino acids, while a complex one may have a thousand. Each amino has to find its place in a correctly-folded chain in order to work. Untangling this knot has proven difficult for computers. Folded proteins are three-dimensional objects, and the number of potential variables is so high that computers are simply inefficient or ineffective.
Humans, however, have an innate, almost mysterious, intuition for pattern recognition and puzzle solving. A computer can only run through every variable, evaluate them all, and try to choose the best one. Puzzle-solving humans don’t do this. We don’t run through every possible version of a puzzle to solve it. We use observation, deduction, intuition, and spatial reasoning skills to evaluate the puzzle, then attempt to find the solution. It’s a vastly more efficient method of problem solving.
The University of Washington’s Center for Game Science decided to tap that power by creating Fold.it, a free game that challenges people to fold protein chains into the most efficient possible shapes. The game begins as a serious of tutorials to introduce the tools and the concepts for protein folding. Each puzzle consists of a 3D representation of a known protein.
The goal is threefold. First, you need pack the protein folds as tightly as possible to eliminate any gaps. Some aminos need to be touching water (hydrophilic) and some need to be protected from water by other aminos (hydrophobic). Thus, the second goal is to place the hydrophilic/hydrophobic aminos in their proper place. Finally, you need to eliminate the “clashes,” which are the places where aminos occupy the same space, or are merely too close together.
The first goal of Fold.it was to see if humans could properly solve protein folding problems. The proof-of-concept trials went well, so Fold.it presented puzzles that were challenging the current automated folding software. People could compete, alone or in teams, for “high scores” by folding proteins in the most efficient manner.
Fold.it had its genesis as an experiment in distributed computing, with people volunteering their computers to aid in the complex calculations needed for protein modeling. The participants would see a screen saver that showed the progress of the automated models, but those watching the screen quickly noticed that the automated modeling was not efficient. That’s when the team created Fold.it to allow people to modify the protein models themselves.
The breakthrough came when the gamers were presented with the Mason-Pfizer monkey virus (M-PMV) retroviral protein, which cause a form of simian AIDS. In an article in the journal Nature Structural & Molecular Biology, the Fold.it team observed that “retroviral proteases (PRs) have critical roles in viral maturation and proliferation and are the focus of intensive antiretroviral drug development work.” Various software models were unable to resolve the structure of the M-PMV retroviral protease. Gamers were provided with a variety of unsatisfactory M-PMV PR models, and then challenged to create a better one. Scientists had been struggling with the structure of this protein for over ten years. When the models were made available on Fold.it, a group of gamers on three continents were able to collaborate and solve the problem in a matter of days.
With the input of human problem-solving, the software modeling suddenly improved. After 600 gamers on 41 teams submitted 1.25 million solutions, the team was able to narrow the field down to 5000 by comparing the models against x-ray data. One team, “The Contendors,” submitted a model that aligned perfectly with this data, and the Fold.it team knew they’d cracked the code.
As the authors write in their Nature article, “The critical role of Foldit players in the solution of the M-PMV PR structure shows the power of online games to channel human intuition and three-dimensional pattern-matching skills to solve challenging scientific problems. Although much attention has recently been given to the potential of crowdsourcing and game playing, this is the first instance that we are aware of in which online gamers solved a longstanding scientific problem. These results indicate the potential for integrating video games into the real-world scientific process: the ingenuity of game players is a formidable force that, if properly directed, can be used to solve a wide range of scientific problems.”
And thus the Fold.it experiment continues. You can join the process as http://fold.it.