Pande and Gruebele then set out with the goal of finding how long it would take the BBAW protein to unfold after a jump in temperature, as might occur if it was present in someone with a fever. This would be hard for Folding@home, as accurately measuring the time it takes a protein to fold had been a tough challenge for computers until that point.
Gruebele explains that back then "people were incredibly happy when their computer just ran one trajectory". He points out that the problem with this was that "if you run just one single simulation and watch it fold, one of several things might happen - it might not fold at all, because you can't run it for long enough, or it might fold to an incorrect structure because your computer force field isn't accurate enough, or it might fold after a specific amount of time, such as five microseconds".
As Gruebele says, this doesn't actually tell you how fast the protein will fold in real life. "When you repeat this," he explains, "different proteins will take different amounts of time - some slower and some faster, and if you only have one computer observation, you don't know whether that five microseconds is just one of the lucky ones that folds really fast."
This is where the power of Folding@home proved itself. Using six months of computer time on Folding@home, Pande was able to run thousands of simulations, and then analyse the speed of the dozen or so proteins that successfully folded. Some were slower or faster than others, but Gruebele explains that "by knowing how many trajectories he [Pande] had run, and what fraction of them had folded, he was able to extract an accurate folding time for the simulation."
Gruebele adds the disclaimer that this doesn't mean that the time he got from the simulation was an accurate measurement, but at least he was actually able to get a time. That's what he was able to do that people had not been able to do before - predict that this protein will take X microseconds plus or minus.
Meanwhile, Gruebele and his team at Illinois performed their experimental measurements, and the two teams then compared their results, which turned out to be surprisingly close. "To be honest," says Gruebele, "it was even closer than I would have ever thought." Folding@home had proved that it could accurately simulate protein folding, with results that could be verified by traditional experimental testing. A paper called 'Absolute comparison of simulated and experimental protein-folding dynamics' detailed the tests, and was published in Nature in 2002.
Gruebele and Pande have stayed in touch since, and Gruebele says he's now in talks with Stanford about using Folding@home to perform computational simulations of how proteins fold when put under pressure. Similar collaborations with other experimental researchers have also been performed since, while others are also in the process of being verified. There's a full page of 63 results papers on Stanford's website at http://folding.stanford.edu/English/Papers, but Pande also says there have been some interesting discoveries that haven't been published.
Pande describes a recent test carried out by Brian Dyer from the Los Alamos National Laboratory in 2008 that was similar to the Gruebele blind test. The findings haven"t been published, but Pande says that they"ve been publicly discussed, and that Dyer "showed that many of the predictions that we made are bearing out to be verified in his most recent experiments". However, there are also some potentially very exciting results from collaborations with experimental labs that have yet to be publicly disclosed.
In less than a decade, Folding@home has been through its teething phase as a small start-up project that needed to be proved and tested, and has now moved on to be useful for medical science. "That's the part I'm probably most excited about," says Pande, "because the work that Grubele and Dyer do are very important for validation of the general technique, but our real goal is to make an impact on something like Alzheimer's."
Good luck, guys. Good luck.
Issue: 133 | February, 2012