Scientists at the U.S. Department of Energy’s (DOE) Brookhaven
National Laboratory and Stony Brook University have been awarded processing
time on a new supercomputer at Oak Ridge National Laboratory to study how
proteins fold into their 3D shapes.
Proteins, which account for one-quarter of our dry body mass,
are made up of different sequences of amino acids. While these sequences are relatively
well understood, the myriad ways in which the amino acid chains fold and curl
are still difficult to predict.
“This is not a matter of just curiosity. Most medicines interact
with proteins and alter their behavior,” explained Stony Brook biochemist
Alberto Perez, a postdoctoral fellow in the Laufer Center for Physical and
Quantitative Biology, one of the lead researchers on the project. “If we want
to design better medicines, we need to know what proteins look like.”
The award, which comes from the DOE’s Advanced Scientific
Computing Research Leadership Computing Challenge, is for up to two million
processing hours on the Titan supercomputer at Oak Ridge. Titan is an upgrade
of Jaguar, a supercomputer that was recently named the sixth most powerful in
the world by the TOP500 ranking system. Some scientists believe that Titan will
contend for the number-one spot when it comes online later this year.
Titan has an unusual configuration: it will run on graphical
processing units (GPUs) in addition to central processing units (CPUs). GPUs,
originally developed by the technology company NVIDIA to render images in video
games, are particularly efficient for repetitive tasks. By re-appropriating
GPUs to solve scientific problems, Titan will become part of a burgeoning new
approach in computational analysis.
Perez’s team will use Titan to run AMBER, a software package
that simulates the effect of different force fields on organic molecules such
as proteins to determine the most likely folding configuration. This method is
a fast and effective complement to conventional protein imaging methods such as
X-ray crystallography or nuclear magnetic resonance imaging.
“With the GPU port to AMBER, we are able to obtain in days in
one single GPU the sampling that we were able to accomplish in several months
of simulation using multiple CPUs,” said Perez.
The trick is finding the right set of conditions to limit
AMBER’s massive search. The team believes they will be successful by focusing
on the “first principles” that govern physical interactions in molecules while
simplifying the representation of electron and nucleus behavior. These
restraints are built into AMBER’s classical molecular dynamics machinery.
The team plans to test their approach in the Critical Assessment
of Protein Structure (CASP), a competition in which groups have just three
weeks to guess at a particular protein’s folds. This is not enough time to run
a true experiment, so most groups try to compare and contrast the protein to
known structures. Having the opportunity to run a simulation on Titan will be a
big advantage.
“Getting into the competition is part of the motivation, but
it’s not the whole story,” said Yan Li, a physicist at Brookhaven. If their
model scores well against CASP’s experimentally derived structure, then they
want to apply their approach to membrane proteins, which are important
therapeutic targets that are particularly difficult to crystallize. “Eventually, we want to make the method more rigorous,” Li said.
“It’s not just inter-institutional. It’s interdisciplinary. We
want to take advantage of the expertise of biology, chemistry, and physics,” Li
said.
Source: Brookhaven National Laboratory