From the New York Times
Herculean Device for Molecular Mysteries
By JOHN MARKOFF
Published: July 8, 2008
A privately financed team of scientists and engineers is nearing completion of a special-purpose supercomputer intended to offer more than a thousandfold increase in performance for complex molecular simulations.
The machine, named Anton, in homage to Anton van Leeuwenhoek, a pioneer in microbiology, is a bold gamble to jump ahead of the most powerful general-purpose supercomputers by as much as a half decade.
It could be used to investigate problems of great scientific interest, like the folding of protein molecules, and in the design of drugs based on the simulated biological activity of different molecules.
The effort is being led by David E. Shaw, a billionaire computer scientist. In the 1990s, Mr. Shaw was one of the most successful of an elite group of technologists pursuing computer-based trading strategies on Wall Street. Several years ago Mr. Shaw, who is also a major investor in Schrdinger, a chemical simulation software firm, stepped away from day-to-day management of his investment firm, D. E. Shaw & Company. He is now chief scientist of D. E. Shaw Research.
The new supercomputer, which is planned to be operational by the end of this year, is described in a technical article in the current issue of Communications of the Association for Computing Machinery, an industry publication.
Anton is massively parallel, with 517 specialized processors working simultaneously. The processors are called application-specific integrated circuits, and in this case their specialty is the calculation of the three-dimensional characteristics of molecules. Despite the publication of a detailed technical description of his computer, Mr. Shaw declined to be interviewed about the project. “At this stage of our research, though, we’ve been trying to communicate our results largely through academic talks and peer-reviewed publications in scientific journals,” he wrote in an e-mail message.
Experimentation in the use of supercomputers to model molecular interactions has been going on for more than a decade, but the field is still largely in its infancy. Simulations of processes like the folding of proteins into three-dimensional structures or the interactions between proteins or between a protein and a drug molecule hold out the promise of advancing science and drug development.
However, each simulation must be validated by experimental scientists in a laboratory setting. Thus one of the principal advantages of increased speed in simulations that now take thousands of hours on the fastest supercomputers is to speed the time to the laboratory.
Scientists said the real value of Anton might not be known until they find out what the machine can do. “Only after Anton van Leeuwenhoek used his microscope did he see protozoa in the pond water,” said Roger Brent, director of the Molecular Sciences Institute, an independent research laboratory in Berkeley, Calif.
The new supercomputer is distinguished from other molecular dynamics computing tools like I.B.M.’s BlueGene/L supercomputer and the Stanford Folding@home distributed computing project in that the machine is designed to simulate a very narrow set of problems on biological processes that take place over a millisecond or longer. Molecular simulations are now done as a series of tiny intervals that may be as short as a femtosecond, one billionth of one millionth of a second, and may last no longer than a microsecond, or one millionth of a second.
By looking at time scales that last several orders of magnitude longer than today’s simulations, the Anton team is hoping to discover new kinds of biological processes that would not otherwise be observable. “If you can do 1,000 times longer, real proteins come into play,” Mr. Shaw said in a technical lecture in 2006 at Stanford describing his work.
Several scientists who are now using general supercomputers said they were eager to see if Anton would produce useful results.
“He’s making a big step forward with this,” said Benoit Roux, a biophysicist at the University of Chicago. Dr. Roux said his research group was using a BlueGene/L supercomputer at Argonne National Laboratory to simulate the behavior of potassium channels in neurons. The group has a three-year Department of Energy grant to use the machine, and it will run simulations continuously, frequently using more than 50 percent of the processors on the supercomputer, he said. Anton would be less flexible than his current system but would provide a valuable brute force accelerator, Dr. Roux said.
Vijay Pande, the director of the Folding@home project at Stanford, which computes its problems by distributing portions of the work over the Internet to graphics processors in the desktop computers of volunteers, said his group had talked about sharing research with the Shaw lab. He noted that both approaches had pros and cons and were potentially complementary.
Anton, Dr. Pande said, “is a sophisticated brute force approach of using a lot of power in intelligent ways.” What remained to be proved was whether Mr. Shaw could complete his project quickly enough to stay ahead of the Moore’s Law pace of computing, which is being driven by inexpensive consumer electronics technologies.