"This provides the accuracy missing in previous calculations,
and it makes bigger systems possible."
Sometimes proteins are like the Wicked Witch. Add water, wait a nanosecond and you can almost hear that piteous whimper. "I'm mel-l-l-ting!" Well, not really. But there may be an audible groan from biochemists as they observe simulations of a large protein or DNA in water. For a few hundred picoseconds of simulation time it looks fine, then the molecule appears to unravel. It's been a troubling problem, but with a series of computations at the Pittsburgh Supercomputing Center, Tom Darden and Lee Pedersen seem to have it fixed.
The problem has been molecular dynamics (MD), computations that simulate and predict how molecular structure changes over time. As computing power has increased over the past 15 years, MD has evolved to become an important part of the molecular biology toolkit. The colorful protein structures in biochemistry textbooks represent molecules painstakingly removed from cells and crystallized so that X-ray crystallography can reveal their three-dimensional structure. These structures have been enormously important in advancing knowledge, but as representations of reality they are analogous to butterflies mounted in a museum showcase. MD makes it possible, in effect, for the butterflies to fly; the static molecular structures are starting data for simulations in a cell-like environment where scientists can observe their movements. The most realistic MD simulations include surrounding water molecules and ions, replicating many of the structural forces acting in the cell.
Nevertheless, trying to account for the interactions among the atoms of the molecule itself with each other and with thousands of water molecules is an extremely complicated computational task. As computing power has increased, it has become feasible and desirable to track how the structure changes over as long as several nanoseconds -- less than a fast eyeblink in human time, but as good as a lifetime in protein biochemistry. Such simulations can take hundreds of computing hours, and the outcome can be disappointing.
A few years ago, Pedersen and Darden encountered protein melting face-to-face. "The protein we were simulating would literally shake itself apart in a couple hundred picoseconds," says Darden, a biomathematician at the National Institute of Environmental Health Science. "This melting behavior is even more pronounced for DNA. Generally, the longer you run, the worse the situation."
With simulations at the Pittsburgh Supercomputing Center, the researchers diagnosed the malady -- simulating electrostatics: the attraction-repulsion forces between atoms that aren't bonded to each other, and Darden devised a cure. He came up with a new method -- "particle mesh Ewald" -- that is fast and therefore has the added advantage of making it feasible to study larger structures. "There's two significant outcomes," says Pedersen, a physical chemist at the University of North Carolina. "This provides the accuracy missing in previous calculations, and it makes bigger systems possible." To further exploit the advantages of this new method, a parallelized version is now implemented on the CRAY T3D at Pittsburgh.
This graphic depicts the molecular structure of a large protein, bovine pancreatic trypsin inhibitor, as determined in particle mesh Ewald simulations by Lee Pedersen and Tom Darden. The color coding indicates oxygen molecules (red), nitrogen (blue), hydrogen (cyan), carbon (white) and sulfur (yellow).
The "backbone" of bovine pancreatic trypsin inhibitor coded according to secondary structure: Random coil (wheat) is tube shaped; helices (aqua) are flat; and sheets (magenta) are arrows. The dotted structure shows the solvent accessible molecular surface.
Researchers: Tom Darden, National Institute of Environmental Health Science; Lee Pedersen, University of North Carolina.
Hardware: CRAY C90, CRAY T3D
Software: AMBER, Particle Mesh Ewald, PME
Keywords: Proteins, molecular dynamics, biochemistry, molecular structure, electrostatics, bonds, non-bonded interactions, particle mesh Ewald, Ewald summation, cutoff radius, biomolecules, counterions, X-ray crystallography, fast Fourier transforms, FFT, macromolecular Ewald summation, bovine pancreatic trypsin inhibitor, H-ras p21, DNA.
Related Material on the Web:
The PSC Biomedical Supercomputing Initiative.
Information on AMBER, including T3D implementation of PME.
Projects in Scientific Computing, PSC's annual research report.
References, Acknowledgements & Credits