INDUSTRY

Scientists at the U.S. Department of Energy's Brookhaven National Laboratory are proposing to use a supercomputer originally developed to simulate elementary particles in high-energy physics to help determine the structures and functions of proteins, including, for example, the 30,000 or so proteins encoded by the human genome. Structural information will help scientists better understand proteins' role in disease and health, and may lead to new diagnostic and therapeutic agents. Unlike typical parallel processors, the 10,000 processors in this supercomputer (called Quantum Chromodynamics on a Chip, or QCDOC, for its original application in physics) each contain their own memory and the equivalent of a 24-lane superhighway for communicating with one another in six dimensions. This configuration allows the supercomputer to break the task of deciphering the three-dimensional arrangement of a protein's atoms -- 100,000 in a typical protein -- into smaller chunks of 10 atoms per processor. Working together, the chips effectively cut the computing time needed to solve a protein's structure by a factor of 1000, says James Davenport, a physicist at Brookhaven. This would reduce the time for a simulation from approximately 20 years to 1 week. "The computer analyzes the forces of attraction and repulsion between atoms, depending on their positions, distances, and angles. It shuffles through all the possible arrangements to arrive at the most stable three-dimensional configuration," Davenport says. The technique is complementary to other methods of protein-structure determination, such as x-ray crystallography -- where the pattern of x-rays scattering off atoms in crystallized proteins is used to determine structure. It will be particularly useful for proteins that are impossible or difficult to crystallize, such as those that control the movement of molecules across the cellular membrane. The high-speed analysis will also allow scientists to study how proteins change as they interact or undergo other biochemical processes, which will give them more information about the proteins' functions than available from structural studies alone. Davenport and colleagues at Stony Brook University will test their application on a QCDOC machine that has been developed for physics applications at Brookhaven by Columbia University, IBM, and the RIKEN/BNL Research Center. To hear more about the potential for using such a machine for studies of proteins, see Davenport's talk during the "Molecular Biology and Computation Session" on Friday, March 26, at 11:15 a.m. in room 510C. This work is funded by the Office of Advanced Scientific Computing Research within the Department of Energy's Office of Science and Brookhaven Laboratory discretionary funding.