Applications of Distributed Computing to Molecular Modeling and Drug Discovery
Thursday, April 10, 1997 - 2:00pm - 3:00pm
Brian Luke (IBM)
A significant programming effort has been applied to porting, or writing, Computational Chemistry programs so that they run efficiently on a multi-CPU architecture; either a shared or distributed memory single image machine, or a network of independent processors. Though these advancements have decreased the elapsed time needed to perform large calculations and have allowed accurate computational procedures to be applied to larger systems, they have not fundamentally changed the way that molecular systems are examined. This talk will present examples of how a distributed, or network, computing model can be used to examine problems in Molecular Modeling and Drug Discovery in a new way. In particular, results will be presented showing how a Parallel Genetic Algorithm efficiently finds the global minimum of flexible polypeptides and how full conformational searches of known inhibitors of the Angiotensin-Converting Enzyme (ACE) can be used to suggest possible pharmacophores. This presentation will also describe how the Bootstrap Method can be used to generate statistically significant results from a relatively small number of molecular dynamics simulations and/or QSAR studies using cross validation as a measure of accuracy. Finally, by distributing a database of possible ligands across multiple workstations, an effective procedure of searching for putative substrates of a particular protein will be described.