The APES Project
Posted: 23 Apr 2013 | 10:23
The picture of a great ape cousin hoarding food at Edinburgh Zoo is deliberately misleading! The "APES" acronym (pronounced "A-PES") actually stands for Advanced Potential Energy Surfaces, and refers to a new project that EPCC is involved in. The project in question is an NSF-EPSRC funded US-UK collaboration that aims to incorporate APES into a range of computational chemistry packages. EPCC's main contribution will be to parallelise software to take advantage of the large-scale compute resources offered by supercomputing clusters such as HECToR and its upcoming successor, ARCHER, as well as NFS-provided resources in the US. This should equip researchers with better tools to advance their understanding of the structure and function of molecules such as, hypothetically, the smell molecule isoamyl acetate (shown), which interacts with simian olfactory receptors to give bananas their irresistible allure.
Potential energy functions, which chemists confusingly refer to as force fields, are crucial in computational chemistry, especially molecular mechanics and molecular modelling. Established force field models implemented in computational chemistry software mostly use non-polarisable fixed point charge approximations. This is done because they are computationally cheap and give reasonable results for equilibrium properties and for homogeneous systems. However they fall short in describing many-body effects, dynamical properties, heterogeneous systems and systems that are out of equilibrium. This represents a major factor limiting the successful application of computer simulations to a variety of grand challenge problems in computational chemistry, biochemistry and materials science.
To get round this one can use polarisable empirical force fields. These force fields allow atom-centred charges to change depending on their environment. They offer clear and systematic improvement in accuracy and are important for future grand challenge applications such as design of environmentally friendly materials, chemical reactions and reactivity, and biological complexity such as protein-drug interactions.
Most major research groups in force field development have recently developed models of polarisable force fields. These use approaches ranging from empirical models, which make approximations by parameterising aspects of the potential calculation, to ab initio calculations that do everything from scratch and from classical to fully quantum mechanical models, all offering different advantages and disadvantages. Major barriers to the uptake of polarisable force field models are:
Existence of a number of competing models
Lack of a canonical or consistent implementation in popular molecular dynamics packages
Inefficient use of high performance large-scale parallel computing resources by packages that do implement these models
AMOEBA (Atomic Multipole Optimized Energetics for Biomolecular Applications), is a prominent empirical polarisable force field developed by J. Ponder et al. and includes polarisable atomic multipoles through quadrupole moments derived directly from ab initio QM electron densities for small molecules and molecular fragments, allowing:
- Replication of molecular polarisabilities and electrostatic potentials
- Fine-tuning of subtle directional effects in H-bonding
- Response to changing or heterogeneous molecular environments
- Direct parameterisation against gas phase experimental data and QM results
- Consistent treatment of intra- and intermolecular polarisation through a physically motivated damping scheme for local polarisation effects
- Bond-angle cross terms
- Wilson-Decius-Cross decomposition of angle bending in-plane and out-of-plane components
- "Softer" buffered 14-7 van der Waals form
AMOEBA is implemented in TINKER and Amber and has been validated to give good results in a number of cases not well described by established force fields. It is also used in the Folding@Home project. However AMOEBA is currently not directly (or easily) implemented in major molecular dynamics packages such as GROMACS, NAMD, or CHARMM. The existing TINKER implementation uses OpenMP for shared-memory parallelism, and does not scale well to more than a few cores. TINKER currently provides the reference implementation of AMOEBA.
So what are EPCC together with the Software Sustainability Institute going to do? Well our contribution to the project will be:
- To provide efficient parallel (hybrid OpenMP / MPI) implementations of algorithms needed to be able to use the AMOEBA force field parameterisation on high-performance computing clusters
- To test and validate these algorithms in TINKER but also to promote interoperability with other packages (Amber, DL_POLY, ONETEP, and Q-Chem) in order to promote uptake.
All of this will be done using an open development process to build a community around those packages that implement AMOEBA and variants of this model, which, in the long term, will make the future development of this force field self-sustaining. This will also make it easier for AMOEBA to be adopted by other software packages thus growing the community that can exploit this method.
This international partnership consists of members from EPCC at the University of Edinburgh, the UK Software Sustainability Institute, the University of Southampton, Daresbury Laboratory and from the US we have the University of California (Berkeley), Rutgers University, Claremont Colleges, Washington University in Saint Louis and New York University. The project will start in mid-April of this year and we shall blog updates as the project progresses.