Before starting GMMX, two files are needed: a structure file in MMX format which is best benerated by PCMODEL, and a command file generated by the GMMXINP program. See the Guide to PCMODEL for notes about how to make the structure file in MMX format. Once you have the structure file, type gmmxinp to run the GMMXINP program to create the command file. Then carefully check your comand file to verify that it instructs GMMX properly; see the GMMX Manual for details. Finally, type gmmx to run GMMX.
All GMMX files are located in /remote/server/gmmx.
GLOBAL-MMX (GMMX) is a steric energy minimization program which uses the MMX force field and operates in batch mode. Its main purpose is to search conformational space and to list out the lowest energy unique conformations thus produced.
In addition to an output file containing the atom connection table and atomic coordinates of the final structures (which can be visualized using the interactive modeling program PCMODEL), a textual summary file called '<OUTPUT FILENAME>.gmx' is produced which lists the energy and Boltzmann distribution of each low energy conformation. (The name within the <> is a user supplied file name.) The file may also include a listing of query operations (PMR coupling constants, distances, angles, and dihedrals) for each structure as well as a Boltzmann averaged summary.
Structures may contain up to 296 atoms total and up to 30,000 structures may be minimized and kept starting from a single input file. On a PC or a Macintosh this is limited to 10,000 structures. The number of structures may be minimized per day using a 386 class PC, Microvax II or MACONTOSH IIx. A Risc-based UNIX machine can be 10-20 (or more) times faster.
GMMX uses two input files, a structure file in MMX format which is best benerated by PCMODEL, and a command file generated by the GMMXINP program. The command file instructs GMMX on which type of search is to be performed and under what conditions. The commands are described in the GMMX manual. This file should be carefully checked to see that it instructs GMMX properly.
It must be remembered that the energies are evaluated in absence of solvent interactions. Although solvent interactions may not be important for hydrocarbons, they can become significant with molecules such as peptides or any molecule which contains many polar functional groups. To treat these types of molecules more effectively, solvent interactions have to be taken into account. A routine to do this is being developed.
Although the conformational searching techniques that follow are unique in their approach, the methods described by the Still group (see iM. Saunders, K.N. Houk, Y-D Wu, W.C. Still, M. Lipton, G. Chang, and W Guida, J. Am. Chem. Soc., 112 p.1419 (1990) and references to previous work cited there-in) were the inspirational source for this work. We also wish to thank Professor Still for sharing with us unpublished work and code that greatly enhanced our routines. We have been beta-testing these routines for a little more than two years now and the results of this testing with a full description of the algorithms is being written for publication.
Conformational space can be searched by GMMX from a PCMODEL-generated structure and command file in five ways:
The Grid Method will systematically rotate each of the selected rotatable bonds (either chain or ring) and minimize the resulting input structure. The rotation increment can be set by the user or the default resolution can be used. For rings, one of the ring bonds is broken and the other bonds are then treated as a collection of chain bonds. After the rotations are done, GMMX then checks to see if the ring can be reclosed without severe distortions and without bad 1,5 interactions. IF these criteria are met, the structure is then minimized. Since the Grid Method is a systematic method, the number of calculations to be done is approximately {the number of bonds} raised to the power of {the resolution}. This number quickly becomes very large and the Grid Method is not recommended except for very limited cases.
The Statistical Methods will randomly select a subset of the rotatable bonds or atoms for movement, followed by checking of the resulting input structure and minimization. These methods depend upon the idea that energy minimization finds local energy minima and that random sampling of the potential energy surface will more rapidly search the surface. In general, the Bonds method gives a more rapid search of wide areas of conformational space, while the Cartesian method gives a finer grain search of more narrow areas. The preferred method is the Mixed method, which alternates between Bond and Cartesian movement.
Molecular Dynamics uses Newton's equations of motion to follow the trajectories of atomic motion. Our implementation uses the Leap Frog Verlet algorithm coupled to a temperature bath. Since the dynamics methods are not energy minimizations, they do not suffer from the local minimum problem. However, since the time steps used in the dynamic simulation are on the order of 1 femtosecond, very long simulations are required to search large areas of conformational space. The dynamics methods have found most use for very large structures where the Grid Method and Statistical Methods are inadequate.