Crowdsourcing Protein Folding with fold.it

I've been working hard on the human-based TSP solver application. I've done some experiments, written some technical reports, and gave a presentation to my research group. The idea appears viable. The rails application is starting to take form, users can make contributions, and I'm thinking about effective sub-problem presentation schemes (an applet for now).

I came across a similar project in the popular news today called fold.it which crowdsources protein folding as a computer game. It was covered on unews, slashdot, technology review, and hacker news among many others. The project appears to be about a year old with some serious backing, stemming fro UW's Baker Lab and Rosetta@home distributed protein folding project. Like Rosetta, the fold.it game is a desktop application that communicates with the internet to download problems and upload results (same core codebase I believe).

The game is pretty damn cool. The interface is simple and slick, and the game itself quickly increases in complexity which I think translates into fun. Each "puzzle" involves manipulating the side chains and backbones of proteins toward minimising the Gibbs energy whilst ensuring the structure is viable. I really like the idea of user manipulation involving both manual and automated tools (jitter, etc.) Problems do not have known solutions, rather an estimation or minimum score (in the training rounds) which has to be obtained before progressing. There are competition puzzles beyond training that I assume do not have this limitation, translating into a free for all optimisation like programming competitions of old (think DrDobbs).

The structure of the site is very similar to my vision for the TSP application, with the important difference that the fold.it game is desktop driven rather than in the browser. The site and game itself are all about the puzzles, promoting both user scores as well as groups of affiliated users. At a minimum, the game provides further verification on the approach toward making difficult problems accessible to be solved by non-domain experts.

The project claims their aim is to locate protein folding savants and exploit them. This is an important general consideration in crowdsourcing applications: locating and rewarding experts. I think a better aim would be to collect statistics from the humans and use the provided solutions as a dataset toward developing better automated techniques for churning much larger and hard proteins. I would also think about how to represent the problem in various different ways to exploit different spatial information and make it less complex for the user on bigger problems. No doubt these are considerations for future work.

It's a cool application, go and download it!