TCAMC Status Report

TEXAS CENTER FOR ADVANCED MOLECULAR COMPUTATION

NSF/ARPA Grand Challenge Group for Computational Biomolecular Design

The existence of a second "back door" to the active site of acetylcholinesterase discovered through computational simulation.

The acetylcholinesterase dimer (AChE), partly shown in the figure, is an enzyme responsible for degrading the neurotransmitter acetylcholine in species from man on down to insects. AChE is a target for many commonly used drugs and toxins, including nerve gas. Among the drugs that bind to AChE are therapeutic agents for Alzheimer's disease, myasthenia gravis, and glaucoma. Recently, our Grand Challenge Group identified the existence of a second, "back door," to the active site creating the likelihood that substrates can come in one door and exit through the other. Now, the Intel Touchstone Delta and Paragon at SDSC will be used to run more extensive simulations that should yield more information about how this "back door" works: what makes it open and close, what can pass through it and what can't.

The computations are being done using the scalable, parallel molecular dynamics program, EulerGromos, developed by the our Grand Challenge Group. Using 256 processors on the Delta, simulations of the full solvated dimer involving 131,653 dynamical atoms, take approximately 20 seconds per time step and require on the order of 50,000 time steps. Access to the Intel Touchstone Delta System, operated by Caltech on behalf of the Concurrent Supercomputing Consortium, is being provided by the Center for Research on Parallel Computation. Access to the Intel Paragon at SDSCis being provided by NSF.

OBJECTIVE: Development of application software using methods for maintaining and improving such software efficiently for a wide range of parallel machines. One goal is to achieve supercomputer performance (and better) on low-cost parallel computers for commercial-grade codes with minimal programmer effort.

APPROACH: Physical models and mathematical algorithms will be developed for Computational Biomolecular Design. Using these algorithms and related software, so-called "hero" problems will be attempted. This will guide the development of both algorithms and software and improve our understanding of what can be resolved computationally (and what cannot). Our approach combines expertise in chemistry, chemical engineering, computer science and mathematics.

Recent work by our group and others on data-parallel programming languages and techniques has resolved many of the core programming issues for multicomputers. The programming languages we have developed require further refinement based on experience gained by implementing sophisticated numerical algorithms we are designing. In addition to continuing development of GROMOS, we intend to port a Brownian dynamics code, UHBD, to a high-level parallel language such as |Pfortran or Fortran D. The use of a high-level parallel language will facilitate development and maintenance of complex algorithms for diverse high-performance architectures.

Examples of parallel algorithms we intend to study include 1. fast summation methods for nonbonded forces in molecular dynamics, 2. more efficient (implicit) time stepping schemes in molecular dynamics, and 3. the use of multigrid techniques to resolve the electrostatic force distributions in the Brownian dynamics model.

PROGRESS: We have progressed rougly one third of our way towards our goals. We have ported several codes to both distributed-memory and shared-memory processors. In this we have developed significant expertise in the software engineering to do this successfully and efficiently. We are using at least three different parallel programming paradigms on a regular basis as appropriate for different applications, but can easily translate between them automatically in most cases. We have done significant simulations of biomolecular systems, such as an over 130,000 atom system involving acetylcholinesterase and water, being done on the Intel Delta using eulerGROMOS, a scalable version of the widely used GROMOS code. Several techniques for solving elliptic boundary value problems via iterative methods have been refined which have a high degree of parallelism. These techniques are being developed to solve as broad a class of problems as possible, but our primary motivation has come from computing the electrostatic potential around molecules of biological significance. These methods will be useful as coarse grid solvers for parallel multigrid methods.

OTHER RECENT ACCOMPLISHMENTS:

A scalable version of the extended system molecular dynamics program ESP has been ported and optimized for the KSR-1.
Comparison of the explicit parallel molecular dynamics code UHGROMOS with a Fortran D version is underway in collaboration with the CRPC at Rice.
Limited amount of quantum effects now incorporated into the UHGROMOS code and being used in simulations.
Solution of three-dimensional electrostatics problems on grids as large as 260 grid points in each direction in a few seconds in the UHBD code, which simulates Brownian dynamics of molecules in an electrostatic field.

FY-95 PLANS:

install and optimize eulerGROMOS on Paragon; begin simulation of entire acetylcholinesterase with Alzheimer's drug bound
install and optimize UHBD on Paragon; apply to predict protonation states of titratable groups in proteins and synthetic "host" molecules
development of adaptive virtual reality techniques for visualizing biomolecular simulations, support for bionautical exploration of computed states of biomolecular systems
complete parallelization of other molecular dynamics codes (CHARMM and ARGOS)
complete integration of parallel preconditioners for solving electrostatics problems into multigrid solvers

960603 JMY