A Scalable Hybrid Approach to Accurately Simulate Biomolecules on SuperMUC
Detailed knowledge about the structural and dynamical properties of biomolecules is essential for Life Sciences, from fundamental research to medical drug design.
Molecular dynamics simulations are a valuable tool that complement experimental results and help to understand them. Molecular mechanics enable simulations of large systems, such as a protein in solution with several ten thousand atoms, up to a microsecond time scale. However, such simulations are by far not accurate enough for tasks like calculating infrared spectra. In contrast, high-level quantum mechanical methods like density functional theory provide the required accuracy, but are computationally limited to much smaller length and time scales.
These issues are resolved by hybrid approaches, which combine the quantum mechanical treatment of a small subsystem with the molecular mechanics description of its environment. These hybrid methods have become a standard tool in life sciences for applications like studying the structure and energetics of enzyme reactions, excited-state properties or charge-transfer processes, as has been acknowledged by the Nobel price in Chemistry 2013 to Warshel, Levitt and Karplus..
Within a KONWIHR-III funded project, a team of scientists from Ludwig Maximilian University, Munich, and Leibniz Supercomputing Centre (LRZ), Garching, combined the program packages CPMD, a well known application for density functional calculations, and IPHIGENIE, an in-house developed program package for molecular mechanical simulations. This combination enables an optimal utilization of HPC platforms. The new approach significantly enhances the accuracy compared to conventional hybrid molecular dynamics approaches, and, furthermore, covers interactions of the inducible dipoles from polarizable molecula mechanics with the electron density. For each of the two program parts, the optimal number of MPI processes and OpenMP threads can be used.
First tests on the HPC system SuperMUC at LRZ show that even for a small input data-set (alanin-dipeptide with 22 atoms as quantum mechanical fragment in a periodic box filled with 2,112 water molecules described by polarizable molecular mechanics), the combined application scales up to 2,048 cores.
In another project, DFT/PMM replica exchange schemes will be implemented, which further extends the parallelism of the program. Several 10,000's of cores can then easily be utilized for rapid conformational sampling of the biological molecule at high accuracy.
Dr. habil. Gerald Mathias
Lehrstuhl für BioMolekulare Optik
Ludwig-Maximilians-Universität München, Germany