Encouraging Results

Since the beginning of 1992, Hauptman's team has worked on the Connection Machine CM-2 (and CM-5 that was on site most of 1992) at the Pittsburgh Supercomputing Center and a CM-5 at Thinking Machines in Cambridge, Mass. They have applied their procedure to 15 molecules ranging in size between 28 and 500 atoms. Most of these structures were already solved, and the calculations were done to see if the new method worked. In all cases, Shake-and-Bake cooked up the right structure, usually in a matter of hours computing time.


Unit-cell molecular structure of gramicidin A as determined by Shake-and-Bake. Gramicidin A is an antibiotic composed of 317 atoms. It acts as a membrane pore in cells, which is why the structure forms a tube along its central axis. To determine this structure originally (accomplished by David Langs of the Medical Foundation of Buffalo) required about 10 years using traditional direct methods. With Shake-and-Bake it took several months. Semi-transparent spheres represent atoms and the short lines represent possible chemical bonds. To arrive at a final, refined structure, the Shake-and-Bake solution is fed into a crystallographic refinement program.

A dramatic success came in July 1992, when for the first time Shake-and-Bake solved a previously unknown structure. The molecule was ternatin, a 110 atom compound similar to cyclosporin, the immune system suppressant used in transplant surgery. Russian scientists had worked for years without success on ternatin, and David Langs, a colleague of Hauptman's at the Medical Foundation of Buffalo, tried half a million possibilities over the previous two years using traditional direct methods. After picking a set of trial structures and assigning parameters, Miller fired up Shake-and-Bake before going to bed one night and in the morning had a value that looked right. Within a few hours, Langs worked out the details of the structure.


Unit-cell structure of ternatin as determined by Shake-and-Bake

Another encouraging success came in 1993 with crambin, a 400 atom protein -- the largest structure ever solved by direct methods. Shake-and-Bake cooked up the right answer in ten hours of CM5 computing. "We have definitely strengthened traditional direct methods," says Hauptman, but he withholds final judgment. "We have yet to solve an unknown structure with 300 to 400 atoms. Once we have done 10 or 15 of those without failure, then we'll be in the position to say this works."

In the meantime, Miller is working on tightening his algorithm and cleaning up the code. "We're trying to get a handle on how many cycles we need for different size structures, and we're developing a clean, user-friendly version so crystallographers around the world will be able to use this."

go back to the main screen