Harvard Medical School

Advancing Structural Biology Research

By Barbara Gibson and Frank Lacombe

Herpes simplex virus replication proteins: Herpesviruses, including herpes simplex virus (HSV) and cytomegalovirus (CMV), encode a DNA polymerase composed of a large catalytic subunit and a processivity factor. The crystal structure of the processivity factor of HSV UL42 bound to peptide is shown.

Jim Hogle is the Edward S. Harkness Professor of Biological Chemistry and Molecular Pharmacology at Harvard Medical School (HMS). A few years ago, Hogle began exploring the potential benefits of a high-performance computating (HPC) cluster for his own research, which he had begun years earlier using one of the large single-platform UNIX systems that dominated molecular research at that time.

“Like many other researchers I came to the realization that for a number of reasons — not the least of which was cost — we needed to migrate from ‘big iron’ to a computational cluster,” Hogle says. “The question was: A cluster of what?”

Postdoctoral fellow David Gohara had joined the Hogle Lab as a protein chemist and structural biologist but soon became fascinated with the computational side of structural biology. When Hogle expressed an interest in exploring the next generation of HPC clusters, Gohara offered to help. After constructing and benchmarking a number of UNIX and Linux test systems, Gohara presented the benchmark results and cost analyses. Based on those head-to-head comparisons, the Hogle Lab selected the dual-processor Power Mac G5 running UNIX-based Mac OS X for their new HPC platform.

“Before we set up the Xserve cluster, people would use their Macs or PCs in the wet lab and then go to a dedicated computer room and physically sit at a workstation to do their calculations on a UNIX platform. Researchers can now do everything from their desktop or laptop computers at the lab, at home, or on the road.”

“A few researchers in the lab were previously running some unoptimized applications on their desktop and laptop Macs,” Gohara says. “Before we set up the Xserve cluster, people would use their Macs or PCs in the wet lab and then go to a dedicated computer room and physically sit at a workstation to do their calculations on a UNIX platform. Researchers can now do everything from their desktop or laptop computers at the lab, at home, or on the road. Physical proximity to the computer room is no longer an issue.”

Meeting the Demand for Computational Research

Researchers in the Hogle Lab use a wide range of structural biology programs — many of which are compute intensive — that are common in the x-ray crystallography, NMR, and electron microscopy communities. Programs that have been developed in biological electron microscopy are especially demanding because one has to store and address as many as 50,000 images at once.

Reovirus Core: Core assembly from Reovirus as determined by x-ray crystallography.

“The capability to do 64-bit processing and addressing is critical because of the way our programs are written and because of the amount of information that needs to be held in addressable memory at once,” Hogle says. The Xserve G5 cluster provides 64-bit support, and the 64-bit libraries included with Mac OS X made it possible for Gohara to optimize the lab’s applications for 64-bit operation. Each compute node in the Xserve cluster is set up with 4-8GB of RAM. Using a job scheduler, memory-intensive calculations are routed to specific nodes, where they will benefit from the extra memory.

“One of the calculations we’re running is something that would have taken a couple of weeks to run on a dual-processor desktop,” says Hogle. “Through optimizations and utilizing the resources available in the cluster, the calculations have been sped up to the point that we can run them a couple of times a day. This means we can run true experimental computing — optimizing not the computational end but the methods.”

The Hogle Lab’s cluster consists of a 14 dual-processor Xserve G5s running Mac OS X. “It was my responsibility to compile and optimize all essential applications on the new systems with a number of new Mac OS X-based workstations in our lab and with the rest of the SBGrid system being utilized in several structural biology labs at Harvard and throughout the country,” says Gohara.

1 2 3

At a Glance

Structural biology research scientists at Harvard Medical School, Harvard University, and Yale Medical School had access to UNIX clusters for their computations — but in order to perform scientific computations, they had to switch from one operating system to another and sit at terminals in a dedicated computer room.

With the introduction of Mac OS X and Xserve G5 systems, there was a good opportunity to create a uniform environment that gives researchers everything they need on one system, as well as the ability to tackle larger problems more efficiently, thanks to the performance of the Xserve’s 64-bit processing power.

Now scientists in more than 38 laboratories, including x-ray crystallography, NMR, and electron microscopy labs, can use their own desktop or laptop computers to conduct computational research at their desks, on the Xserve G5 cluster, or on the road. What’s more, the benchmark results indicate performance increases of anywhere from three to ten times over other benchmarked computing resources.

“The IT group said ours was the easiest cluster installation they’d ever done. One of the Apple engineers came by to help configure the cluster and it was up and running in under three hours. That’s amazing.”

— David Gohara