The science of Folding@home

Distributed computing
As you can see, simulating protein folding demands a colossal amount of computer time, even on large-scale supercomputers. However, Scheraga points out that this is "the problem that Vijay Pande has solved" with the Folding@home project. By using the spare clock cycles of CPUs and GPUs on Internet-connected computers all over the world, Folding@home now has a tremendous amount of processing power. It's a concept that's commonly known as distributed computing, and it's perhaps most well known from SETI@home, which uses distributed computing power to look for aliens.

Scheraga doesn't use Folding@home for his own work, but this is mainly because he likes to have control over the process. "I just don't know how my code will function on other computers," he says. "I've collaborated with lots of people in my years in science, and I've always found that such collaborations don't lead to results in a reasonable amount of time."

What's in a work unit?
So what's actually in a Folding@home work unit? The work varies massively between work units, but your computer will usually only carry out a very small part of a folding process. In fact, the founder and director of Stanford's Folding@home project, Vijay Pande told us that a CPU client with a slower protein might only calculate "one millionth of the process".

When we talk about the speed of a protein, we're simply talking about the time that a protein takes to fold. Pande points out that proteins fold on a timescale measured in microseconds or milliseconds, and that a Folding@home work unit could represent "somewhere between a nanosecond and a microsecond of overall dynamics". As such, Pande says that a work unit processed by the GPU client "could get close to folding a whole process on a very fast protein".

Does Folding@home work?
It's worth noting that the concept of having this level of processing power available for protein folding was unprecedented until Folding@home started in 2000, and computational protein simulation was then (and still is) a new science. How could Stanford prove that it could yield reliable and useful data? As any medical expert will tell you, the key to proving your research is often found in double-blind tests. In a medical trial, for example, a double-blind test would mean that neither the patient nor the person administering the drug know whether the patient was being given a placebo or the genuine medicine.

This is what Folding@home needed to prove to the scientific community, and Martin Gruebele, professor of chemistry, physics, biophysics and computational biology at the University of Illinois, was happy to be the other "blind" partner. Gruebele had a research lab for experimental protein folding tests, so he and Pande could see how both computational and experimental results compared.

"Vijay and I happened to attend the same conference at an American Chemical Society meeting," Gruebele explains, "and he was just getting his Folding@home computer system running, so we said let's do a double-blind study - we're not going to tell you the experimental numbers, and you're not going to tell us what results you're getting from the simulation. Let's run both projects for a while in parallel, and see how they compare."

Gruebele points out that the term 'double-blind' should be in quotation marks, as there were no rigorous controls on the procedure, but neither scientist honestly knew the other's results until they were compared. The protein used for the test was based on an artificial protein designed by Barbara Imperiali's research group at the Massachusetts Institute of Technology. Called BBA (one of a series of proteins called BBA, BBB, BBC and so on), the protein formed a part of what's known as a zinc finger.

"We looked at these sequences and decided that one of them could be a very good small protein for a direct comparison between computer simulation and folding," says Gruebele, "because it's very small, it might fold quickly." Gruebele's lab also decided to add tryptophan to the protein to make it fluoresce under the lasers in the experimental lab. As the abbreviation for tryptophan is W, the protein used in the test was called BBAW.