"If the protein tried every possible arrangement until it found the one with the least energy, folding would take longer than the age of the universe," says Harold Scheraga, the Todd Professor of Chemistry emeritus, who has been studying protein folding since the 1960s.

The protein folds, unfolds and refolds hundreds, perhaps thousands, of times until it arrives at the shape with the least potential energy -- that is, where the positive and negative charges in the molecule are as close to one another as they can get. It finishes in a few thousandths of a second, which means the folding process is far from random.

Theoretically, a computer could calculate every possible configuration and choose the one with the lowest energy. But that would take even longer for a computer than it would for the protein. So far, Scheraga says, computers have been able to simulate only the first nanosecond of the few milliseconds of the folding process. So programmers make guesses. Some use the "Monte Carlo" method: pick a shape, change it in some way and test to see if that gets you closer to where you want to go. Others compare random shapes with a database of protein shapes.

Scheraga's research associates, Jooyoung Lee and Adam Liwo, have had success with a method called conformational space annealing. The computer randomly chooses about 20 shapes out of the thousands possible, which become "seeds." Each seed is twisted into several close variations; the ones with the lowest energy become new seeds, and the process is repeated until no lower-energy shapes can be found. Lee and Liwo have obtained the structures of proteins containing 50 to 75 amino acids in about 12 hours of supercomputer time.

Scheraga and his students also examine intermediate steps in the folding process in the laboratory. They create mutated versions of a protein that stop somewhere along the way, then use nuclear magnetic resonance spectroscopy to find the structures of those partial folds.

David Shalloway, the Greater Philadelphia Professor in Biological Sciences and chair of the Section of Biochemistry, Molecular and Cell Biology of the Division of Biological Sciences, heads a research group that takes a statistical approach. Not only is it impossible to simulate the motion of every atom, he says, but it's not necessary. To predict the folded shape of a protein, he samples all possible shapes, zeroes in on a group that seems to have the lowest average energy and narrows from there.

Shalloway also studies the ways proteins refold or deform during chemical reactions, such as when a drug binds to a protein or cell receptor. He finds that substructures of the protein usually stay together, and he can compute the average movement of a large group of atoms rather than trying to simulate the movement of each one separately.

The researcher doesn't have to depend on computer simulations to identify the protein made by a particular gene; in most cases the protein can be synthesized and its structure determined in the laboratory.

But, Scheraga emphasizes, "Prediction is not the major goal, but to understand how the folding process occurs."