May 092008

By now many of the nerds in my audience have heard about Foldit—the server seems to be wanged, so clicking on that link may not do anything for you. The idea behind Foldit is to use the processing power of human beings to address the protein folding problem in the context of a game. Despite many years of trying, predicting the fold of a protein and moving a random coil into this shape is a task that computers typically find very difficult. Distributed computing efforts, like Vijay Pande’s folding@home or Rosetta@home, are one way of addressing this problem for some limited cases. Foldit, from David Baker’s group, is conceived as a way to train the ROSETTA program to get better at moving structures to their ultimately predicted shape. This may sound complicated, and the system isn’t perfect, but almost anyone can play and it’s surprisingly fun.

The game is relatively simple to use, if a little flaky. You are handed a peptide chain with secondary structural features already formed. You can move these motifs into the form of a packed structure by manually moving them around, by letting ROSETTA wiggle them freely, or by giving ROSETTA general directives about where to move the elements during this wiggling. Directing the automated movements is pretty seamless, though clashes sometimes have a way of causing the whole protein to blow up. Many of these clashes can be resolved by “shaking” the side chains, also a relatively seamless automation.

However, your manual control is limited and occasionally awkward. Sometimes, Foldit just won’t move a side chain where you want it to go, and at other times it will persistently rotate the wrong angle. As I noted above, the secondary structural elements are pre-set, which is okay for helices but seems a little off in the case of strands. Also, many of the loops are much more rigid than expected. Many of these problems could be solved by context-sensitive menus that allowed you to select and rotate a particular bond angle; this could also help with the axial rotation of β-strands, something I very much wanted to do in two of the puzzles. I have yet to see hydrogen bonds or charge-charge interactions between side chains, and indeed the game has comparatively little information about side chains generally. In default mode, unless they clash, they’re visible only as little nubs with a faint glow to tell you whether they’re hydrophobic or not. A lot of this may be simplification on the part of Foldit to prevent “impossible” conformations, but this is a limitation the creators should explicitly try to avoid. Avoiding what it feels to be impossible may be one of ROSETTA’s flaws, after all.

Playing Foldit is an interesting experience, and the puzzles are fun and challenging without requiring any detailed knowledge about biochemistry. People who do know something about biochemistry, however, may find themselves frustrated by their inability to make moves that seem sensible. And, the simplicity means that people who don’t know any biochemistry probably won’t learn anything lasting from Foldit. The controls are a little wonky and at times seemingly unresponsive. Still, this is just the open beta version, so hopefully some of these minor annoyances will be fixed going forward. Once the furor dies down (in about a week or so), you should head over to the site and see if you’re interested. The game is free to download, but requires an internet connection.

In related news, an article by Baker that I covered earlier concerning the use of ROSETTA in combination with directed evolution to produce relatively active enzymes has been published in this week’s edition of Nature. It’s an interesting article, and definitely worth reading if you have an interest in protein engineering.

  One Response to “You’re like a tiny god!”

  1. you fool! I will pwn joo with my mad 5h337 makin' skillz!!

Sorry, the comment form is closed at this time.