Last author

We have long known that a protein's amino-acid sequence dictates its three-dimensional structure, but the structures of most proteins remain unknown. In most cases, the correct structure exists when the molecule is in its lowest energy state. But finding that state is a huge challenge because of the large number of possible conformations. On page 259, David Baker, at the University of Washington in Seattle, and his colleagues describe a computational approach to refining protein structure models that provides highly accurate predictions. Baker explains how thousands of home computers made this work possible.

What part did the home computers play?

Predicting protein structures requires a vast amount of processing power. A couple of years ago, we started a distributed computing network called Rosetta@home. This takes advantage of any spare processing power available in the computers of volunteers while they are online. About 160,000 people are now signed up worldwide, although probably only about 20,000 are online at any given time. Their contribution allows us to do things that we simply couldn't do using our in-house computing resources alone. Rosetta@home is equivalent to a reasonably sized supercomputer, or about 62.7 teraflops.

What else contributed to this work?

A community-wide experiment called CASP (critical assessment of techniques for protein structure prediction) that tests how well current structure-prediction methods work. It has the flavour of a competition because everyone wants to see how well their method stacks up. I met co-author Randy Read, from the University of Cambridge, UK, at the CASP7 meeting in Pacific Grove, California. One of our predictions on a solved-but-unpublished structure was highly accurate. While at the meeting, Randy used our prediction and a program he had written named Phaser, and solved the X-ray crystal structure of the protein.

What is your paper's take-home message?

That you can combine conventional experimental methods such as X-ray crystallography with current protein-prediction methods and solve structures. And that there are cases in which a protein's three-dimensional shape can be accurately predicted using only its amino-acid sequence.

What is your favourite aspect of the project?

It's been fun explaining to people how their computer time is being used and the possible implications of the project — for example, in helping to develop HIV vaccines, anti-malarial agents and gene therapy.