Despite advances in computer technology, computing in macromolecular crystallography keeps pace in its demand for CPU power. Improvements in CPU speed, together with advances in computing methods that depend on it, often translate into the possibility to solve structures that would otherwise require additional experiments. Programs for data-reduction, molecular replacement programs employing multidimensional searches on a grid in real, Patterson or reciprocal space, and phasingand refinement programs currently have, among others, the highest requirements for CPU power.
Parallel computing, which is possible due to the inherent parallelismof crystallographic algorithms, extends the range of problems in macromolecular crystallography that programs can be applied to, and can significantly reduce the time required for going from a dataset toa refined model.
For the algorithms mentioned above, speedup of calculations due to parallel execution on multiprocessor computers is possible with eitherthe OpenMP or the MPI programming interface. Preliminary results concerning parallelization of crystallographic programs with OpenMP were reported in[Diederichs (2000) Computing in macromolecular crystallography using aparallel architecture. J. Appl. Cryst. 33, 1154-1161; online athttp://journals.iucr.org/j/issues/2000/04/00/he0253/he0253.pdf]
The present project is targeted to the parallelization and applicationof molecular replacements programs using grid searches, and to the parallelization and application of newly developed phasing programs.
In the case of molecular replacement programs, the use of the HitachiSR8000-F1 supercomputer will have an effect similar to a reduction ofthe dimensionality of the search problem and will therefore allow forcrystallographic computations that have previously been deemedimpossible due to constraints in CPU speed, resulting in improvedprospects for structure solution, including investigations instructure-based drug design.
For the molecular replacement and other phasing algorithms, the use ofthe supercomputer will allow the mathematically adequate maximum-likelihood target to be computed instead of its less well-foundedapproximations used hitherto, which will ultimately result inimproved quality of X-ray structures.