What is EGO?

EGO [10] is a program to compute molecular dynamics trajectories. It is written in the programming language C [28]. EGO runs as a massively parallel program with high efficiency on a PowerXplorer (Parsytec) and CC (Parsytec) parallel computer using PARIX[34]. To be highly portable EGO uses PVM [17] and MPI [12] on workstation clusters, the Cray-T3D, Cray-T3E, the IBM-SP2 and other parallel computers. Additionaly EGO can be compiled to run sequential and, thus, may run on any UNIX workstation or Windows NT/95 system.

We have used EGO to compute trajectories for molecular systems containing more than 35,000 atoms with extended atoms (i.e., non-polar hydrogens are handled implicitly in terms of special atom types) or with explicit hydrogen atoms. EGO uses a modified Verlet integration scheme (see Chapter 6). Input files to EGO consist of Brookhaven Protein Data Bank PDB files for the atomic coordinates, and X-PLOR-compatible PSF and parameter files for topology information and force constants.

Molecular dynamics simulations [30,40,32] have evolved in codes such as CHARMM [3,4] and X-PLOR to model motions in small molecules, proteins, and nucleic acids in order to better understand molecular structure and function. Compared to EGO X-PLOR [6] is a more extensive package for macromolecule structure determination and refinement written by Professor Axel Brünger now at the Departments of Biophysics and Biochemistry at Yale University. X-PLOR has molecule structure manipulation capabilities and other useful features which complement the computing power of EGO for molecular dynamics (MD) work.

Because EGO and X-PLOR share common data file formats, it is important to understand what capabilities they share and how they differ. A more detailed comparison between them is contained in Appendix B.

In molecular dynamics calculations the equations of motion of atoms in molecules are solved by numerical integration. Electrostatic and van der Waals interactions represent non-bonded forces between atoms, and bonded interactions between (bonded) atoms are represented by stretching, torsion, and stearic hindrance potentials. The computational effort of the short range forces increases linear with the number N of atoms and is for sufficiently large systems (N > 100) negligible compared to the computational effort caused by Coulomb interaction, which increases with N². To reduce this huge computational effort we developed a method which combines a Fast Multipole Method (FMM) [19,20,29] and a Multiple Timestep Method [37,44,23] for rapid, yet sufficiently accurate evaluation of Coulomb interactions. The FMM is based on a multipole expansion of the Coulomb potential to a given order for a hierarchical subdivision of space. Rather than to use a cubic subdivision of space -- as most implementations of the FMM do -- we choose a Structure Adapted [33] decomposition method. This method takes advantage of knowledge about structural and dynamical features of the biomolecules and helps to reduce the computational effort. The Multiple Timestep Method is based on the fact that the influence of far separated atoms varies slowly with time and, therefore, the contribution of these interactions can be evaluated less frequently. The combination of these algorithms, which is implemented in EGO we termed FAst MUltiple timestep Structure Adapted Multipole Method (FAMUSAMM) [9].