Computing New Compounds

When IBM's Watson computer destroyed two Jeopardy champions in February, it revealed how much a modern supercomputer can understand. In the world of drug discovery, however, today's powerful computers go even farther. They create.

The pharmaceutical industry still faces extremely long odds of bringing any one compound from concept to the pharmaceutical counter. "It's a critically important but horribly inefficient process," says Ray Bergan, codirector of the Center for Molecular Innovation and Drug Discovery at Northwestern University. But new molecular simulations at Argonne National Laboratory could save years of expensive lab work and help produce new drugs more quickly.

To identify candidate drug compounds from the millions that chemists could create, scientists have simulated how various molecules can fit into an enzyme's binding pocket. By looking at a picture of a protein and a molecule that binds it, scientists can draw in all the chemical bonds that make the molecule stick in the pocket, says structural biologist Andrew Binkowski of Argonne National Laboratory. "The mystery is how you can get the computer to understand this as well as we can see them," he adds.

Since the physical forces that draw together a molecule and enzyme are well understood, in theory, researchers could program computers to calculate them and add them up to simulate the fit. In practice, however, these calculations are too complex even for most supercomputers. To save valuable computing time, researchers approximate these forces, creating a source of error, Binkowski says.

Binkowski and his Argonne colleagues eliminate the guesswork. They use advanced robotics, a high-energy x-ray beam and fast supercomputers to quickly determine a target protein's structure. Then they use their IBM Blue Gene/P supercomputer, which can blaze through calculations at an astonishing 557 teraflops—or 557 trillion mathematical calculations—per second. This computer and accompanying software move virtual versions of candidate drug compounds around the virtual protein to find those that fit best. "It's almost like a little game, like Tetris," Binkowski says. The extra computing power lets them screen a million compounds in a matter of days, vastly improving the odds of finding a good drug candidate.

The researchers have already scored some hits, including a virtual compound that binds tightly to a protein called transglutaminase, which could be important in Alzheimer's disease. Surprisingly, this compound attaches to the target far from its binding pocket. Binkowski adds that they've found some "halfway decent inhibitors" of key enzymes in anthrax.

Supercomputers could also help speed drug development by making virtual chemical modifications to a compound, then testing how well it binds the target protein. "The promise is that you won't have to go to the wet lab until you've already designed the perfect drug on the computer," Binkowski says.