By Salvatore Salamone
May 12, 2005 | Researchers at the IBM T. J. Watson Research Center, along with a colleague from Wabash College, have published the first scientific results based on molecular dynamic simulations run on IBM’s powerful supercomputer, Blue Gene/L.
Using the processing power of a half-rack of a Blue Gene/L system, the group extended the simulation time for the molecular dynamics over previous efforts by about a six-fold.
Researchers conducted a 118-nanosecond molecular interaction simulation of rhodopsin embedded in a bi-layer systePm containing cholesterol and two different lipids. Previous simulations of much simpler biological systems (e.g., bio-layer systems without the cholesterol) typically modeled only 18ns of interaction.
The results were published in Journal of the American Chemical Society (Role of Cholesterol and Polyunsaturated Chains in Lipid-Protein Interactions: Molecular Dynamics Simulation of Rhodopsin in a Realistic Membrane Environment, J. Am. Chem. Soc., 127 (13), 4576 -4577, 2005.)
Commenting on the reported research, Ajay Royyuru, senior manager of the Computational Biology Center at IBM T. J. Watson Research Center, called the scale of computing used unprecedented. “Problems have existed, but they’ve been unapproachable with existing computers,” says Royyuru. “This gives [researchers] more insight into the actual dynamics taking place.”
Extending the interaction time and the complexity of the systems modeled, provides a much closer to real-life picture. In nature, biological membranes usually contain multiple lipids, cholesterols, and proteins. And experimental results suggest that organisms vary the lipid composition of their membranes to modulate protein function. One example is retinal rod membranes, which are rich in polyunsaturated lipids and cholesterol interspersed with the photoreceptor protein rhodopsin. Understanding the detailed interplay among these components is critical to understanding the biology of that system.
Frank Suits, a computational biology researcher at the Watson Research Center and one of the paper’s authors, notes an important aspect of this type of supercomputer modeling is shedding further light on laboratory experimental results. Comparing simulation to a kind of “virtual NMR” Suits says, “The biological relevance is that a researcher can measure dynamics of molecules and aggregates in the lab and use the simulation’s atomic layer details, which cannot be seen in the lab, to get a better understanding of the molecular interactions.”
This is the first paper done with Blue Gene/L hardware. Suits notes that a number of past papers were based on research done using Blue Matter, which is application software (developed for Blue Gene) and used to run simulations of protein dynamics. However, these earlier efforts were run on other IBM server hardware and not on Blue Gene/L.
The earlier works include a folding simulation study of a beta hairpin (based on about 5,000 atoms) and research that looked at lipid bi-layer interactions (bases on about 12,000 atoms). Building on some of this earlier work, “the next stage was to look at lipid bi-layer with cholesterol,” says Suits.
In this latest effort, more than 43,000 atoms were included in the molecular dynamics calculations, “require[ing] much larger compute power,” says Suits, “and we had to go much longer than 20ns [of previous efforts] to get details.”
Investigators use lab experiments to try to understand what role cholesterol plays in this rhodopsin, lipid, and cholesterol system. “The cholesterol molecule does not touch the protein, but it is involved,” says Suits. Observations in the lab indicate rhodopsin forms into an hourglass shape and the cholesterol’s distribution complements this hourglass shape.
Simulation also yielded this hourglass shape, agreeing with what was seen in the lab, and building researchers’ confidence that the modeling is indeed matching the real-world interactions. Researches can now look at the simulation results to explore how the various elements move in the entire the interaction. And as a result, the authors (in the JACS paper) suggest a molecular-level mechanism for the experimental finding.
Besides Suits, the paper’s other authors are Michael Pitman and Alan Grossfield, both of the IBM T. J. Watson Research Center, and Scott Feller of the Department of Chemistry at Wabash College. l