Computational folding of proteins moved in this year from qualitative to quantitative folding and analysis. The rule-based model of the topological interaction of chemical components in typical proteins that has been under development for several years through a collaboration with David Rawn at Towson State University was significantly improved by eliminating three-body interactions. All interactions must now occur with only water between the atomic components. The recognition for the need for this simple change permitted the development of a very powerful conceptual model of folding in which the proetin is drawn on a sheet of paper. Folding then consists of all the actions which can go on without tearing the sheet of paper. Quantitative measurement of the degree and accuracy of the folding of a protein is achieved by using CHARMM. It became reasonable to do this measurement when the protein structures qualitatively took on the folding pattern of the NMR or X-ray crystal structure. The two Helix protein ROP and cytochrome b562 were calculated to within 4 ? of the respective structures as determined by NMR. The rule-based topological model was then used on the NMR structure to see if it maintained the correct structure. These typical helical structures now stay within 1.5 angstroms of the NMR starting structure. THE problem which remains in the folding simulations is to improve them from approximately 4 angstroms to something better than 1.5 angstroms. The computationally intensive portions of the rule-based folding are achieved by the DGEOM distance geometry program from the Quantum Chemistry program Exchange (QCPE). The simulation of the folding of ROP takes a single processor about 30 days and cytochrome b562 takes about 100 days. Using as many as 18 processors in parallel can reduce computing time. An attempt to reduce the manual work involved in running a large number of processors at many different geographical sites has led to a collaboration with Richard Freund at the Naval Research and Development Center for the use of SmartNet. The use of individual workstations, main frames and super computers can be made far more efficient by having SmartNet decide when and where computational tasks are to be performed. SmartNet is a general meta-computational tool which can be used by man problem areas and has been implemented at the NIH since January 1995. The topological modeling of protein folding contains really three components: the rule-based constraint determination (Lfold), the distance geometry modeling (DGEOM) and the energy refinement (CHARMM). Distance geometry and energy refinement are being combined into a more efficient vehicle through collaboration with Richard Judson at the Sandia National Laboratory. His program, CCMD, combines CHARMM energy potentials with distance geometrical constraints.
