Biological and Environmental Research

The similarity of this simulation to observed conditions shows that the coupled ocean/atmosphere general circulation model faithfully captures the large-scale patterns of low-frequency climate variability. This pattern is associated with changes in winter temperature and precipitation over much of North America.

Computational simulations are now playing a crucial role in biological and environmental research, particularly in the areas of climate modeling and genomics.

NERSC's high performance computers are being used by researchers nationwide for global climate change simulations and other atmospheric studies, as well as investigations of pollutant formation, energy use, and other environmental issues.

A team led by the National Center for Atmospheric Research (NCAR) used the new high-resolution Parallel Climate Model (PCM) to produce a climate simulation for the 1990s which will be used as the control experiment for future climate change scenarios. The NCAR/NERSC collaboration has achieved the fastest climate modeling performance in the nation. Using 256 processors on NERSC's Cray T3E, it takes less than a half hour to simulate one model year with PCM; the same one-year simulation would take more than 10 hours on the older Cray C90.

Collaborators at NERSC and the National Oceanic and Atmospheric Administration's Geophysical Fluid Dynamics Laboratory (GFDL) are adapting GFDL's Modular Ocean Model (MOM) so that it will run efficiently on massively parallel computers for high-resolution, century-long ocean simulations. In the first phase of the project, portions of the code are running 15 to 50 times faster.

Another GFDL team working with an atmospheric general circulation model has found that increased vertical resolution produces far more realistic simulations of winds in the tropical middle atmosphere. This important discovery opens up the possibility of alleviating one of the most serious and widespread problems in atmospheric simulation--the absence of the Quasi-Biennial Oscillation (QBO).

Researchers at the Scripps Institution of Oceanography are working on a coupled ocean/atmosphere model to determine the predictability of century-long precipitation and temperature patterns. At various DOE laboratories, scientists are making seasonal hydroclimate predictions for the western U.S.; developing a global tropospheric/stratospheric model to assess the impacts of energy-related emissions on tropospheric ozone; and investigating the effects of albedo (atmospheric reflection of light) due to sulfate aerosol emissions. A team of researchers from NCAR and academia are using simulations to study the effects of gravity waves on the atmosphere.

The Human Genome Project is producing an enormous database of amino acid sequences. To understand what they mean, we must understand how amino acid sequences result in protein structures and functions. Computational biologists are using simulations to predict a protein's three-dimensional atomic structure--that is, how the protein folds. Accurate computational prediction of protein folds, coupled with comparative modeling and ab initio predictions, could significantly accelerate the interpretation of genome data.

Several different approaches to protein fold prediction are being tested on NERSC computers. One method, which approaches folding as a taxonometric and statistical problem, uses a neural-network-based expert system to assign proteins to folding classes. As a large-screen filter for predicting gross fold topologies, this method has an advantage over detailed sequence comparison, because sequences belonging to the same folding class can differ significantly at the amino acid level.

The global optimization approach has successfully simulated the protein structure of the -chain of the 70-amino-acid protein uteroglobin. Shown here are the predictions from crystal structure (left) and sequence (right).

A second approach--a global optimization strategy for predicting protein structure--is based on the theory that proteins adopt the structure with the lowest free energy level. Therefore, finding the location of the lowest energy level, or the global minimum, helps identify the fold structure. The global optimization approach has successfully simulated the potential energy surfaces of small homopolymers, homopeptides, and -helical proteins, including the 70-amino-acid protein uteroglobin.

Even though the process of protein folding occurs in milliseconds, that is too long for the molecular dynamics of the full process to be simulated on the current generation of computers. An ingenious new way around this problem exploits the fact that a small protein fragment can be unfolded at high temperatures in nanoseconds. After simulating the all-atom molecular dynamics of the unfolding process, researchers were able to perform a series of short simulations of protein refolding, starting from the discrete transition states determined by the unfolding simulations. The refolding simulations showed a surprisingly good agreement between high- and low-temperature transition states, as well as experimental results under physiological conditions.

In addition to protein folding, computational biologists are exploring the fundamental dynamics of other biological processes. For example, one team of researchers performed the first molecular dynamics simulation of carcinogen-damaged DNA replicating and mutating in the presence of an enzyme. This 12,000-atom simulation demonstrated the mutagenic effect of benzo[a]pyrene, a common pollutant found in automobile exhaust and tobacco smoke.