PNNL researchers earn top honors at Supercomputing conference
December 03, 2008
Interactive program that can search through DNA wins contest
RICHLAND, Wash. –
DNA sequencing is easier than ever, but the amount of data to be analyzed is piling up. An award-winning computer program now shows that genome sequence analysis can be made interactive and intuitive, helping researchers find hidden relationships in massive amounts of data.
Researchers from the Department of Energy's Pacific Northwest National Laboratory captured “Best Overall” for their entry at the Supercomputing '08 High Performance Computing Analytics Challenge in Austin, Texas, on November 20.
In the competition, scientists were judged on solving real world problems using comprehensive computational approaches, large data sets, and high-end visualization technology to display results — which means it had to look good and be easy to use.
PNNL’s Chris Oehmen led a multidisciplinary team composed of Scott Dowson, Chandrika Sivaramakrishnan, Justin Almquist, Lee Ann McCue, Bobbie-Jo Webb-Robertson, and Jason McDermott to the win. The team used resources at PNNL and at EMSL, DOE's Environmental Molecular Sciences Laboratory on the PNNL campus, to develop the interactive program.
Past finalists have been in areas as varied as orthodontics, atomic energy, and music classification. PNNL's winning entry in genomics combined multiple databases, analysis software, and a home-grown "visualization technology" called Starlight that presents data in unique visual patterns and allows users to interactively explore them.
“Our entire team is thrilled that we won,” Oehmen said. “It's an honor to be a part of this international competition. We could not have completed the challenge without the support of our sponsors at the Department of Energy, the National Science Foundation, and internal investments from the Pacific Northwest National Laboratory.”
A common problem for genomics researchers, said Oehmen, is that desktop computers often can't handle the volume of calculations needed to analyze many genomes at once. At the other end of the spectrum, high performance computers often limit researchers' ability to guide the analysis along the way.
“We wanted to demonstrate that high-performance computing can be integrated into an iterative workflow because this is the way biologists really work,” Oehmen said. “It was the MeDICi middleware that really helped us pull the various data, analysis, and visualization together."
In genomic studies, computer programs compare DNA sequences of different living things to find shared proteins or uncover the function of a mystery protein, generating ideas that can then be tested in laboratory experiments. This interactive program gives laboratory researchers a place to start in looking for proteins and genes with interesting functions.
Oehmen demonstrated that their interactive program and high computing power could explore the complement of proteins found in an organism, allow them to focus on a protein that intrigued them, and investigate its possibilities.
Browsing through all the proteins in various Shewanella bacterial species, the team noticed more proteins than expected with telltale iron-detecting components.
"We thought, there are a lot of iron-sensing proteins here. What are they doing?" said Oehmen.
As it happens, many species of Shewanella have the ability to transfer electrons to an electrode, thus forming a simple biological fuel cell, an alternate means of generating energy. Iron is involved in this activity, so the team decided to identify proteins that may help the bacteria sense the iron and form a biofilm on an electrode.
Starting with a known, non-Shewanella protein that senses iron, the program allowed researchers to guide the search for similar proteins out of 42,000 proteins from 10 Shewanella species. After rounding up about 550 possible iron-sensing proteins, the researchers switched gears and determined which of these might also be involved in biofilm formation, based on other criteria. Ultimately, the team zoomed in on one protein that had potential roles in both activities. In addition, some of the species had two copies, suggesting those species might have some sort of biofilm advantage.
"Letting users find this sort of information interactively is the main motivation for this work and for the visual representations we have chosen," said Oehmen.
The visual representations of the data included colorful "graphical clusters" that looked like pie charts to the untrained eye, and other images that looked like stars connected through space. In the Shewanella example, the star-graph clued the researchers into the presence of the extra protein copies.
"Presenting data visually can let important information rise to the top," said Oehmen.
And there at the top, the secrets buried in DNA data just can't stay hidden.
A video demonstration of the program can be found at Fundamental & Computational Sciences Awards.
Software developed by PNNL that were used for the demonstration included SHOT, a sequence analysis algorithm that transforms protein sequences into sets of features to identify homologous pairs using a support vector machine; ScalaBLAST, an algorithm for performing gene and protein sequence analysis; Starlight, now commercially available from Menlo Park-based Future Point Systems, Inc, an information visualization application to perform advanced interactive visual analysis. Bringing them together was MeDICi, a "plumbing" software technology developed as part of PNNL’s Data Intensive Computing Research Initiative that supports the integration of applications, data and computing resources.
Tags: Energy, EMSL