A small team of researchers at Argonne National Laboratory is quietly developing a platform that could usher in the much-anticipated era of grid-enabled bioinformatics tools. Called the GADU (Genome Analysis and Database Update) system, the automated sequence analysis pipeline architecture is the first genomics-based application to run on the US Department of Energy’s Science Grid — an ambitious project that began in 2001 with the goal of linking the computational resources of the entire DOE lab system.
The ANL bioinformatics team, led by Natalia Maltsev, has expert guidance in navigating the unfamiliar grid landscape. Ian Foster, a grid computing pioneer and co-developer of the open source Globus toolkit, heads up a team of six Globus developers who are working closely with Maltsev’s group to bring GADU online.
For Globus, which was first released in 1998, biology is new territory, said Foster.“When we started, the primary user base was in the physical sciences,” he told BioInform, “but increasingly we’re seeing interest from the life sciences in the use of Globus as a technology for data federation, federation of computing resources, and other such things.”
The Globus/ANL team recently completed a proof-of-principle project that ran the GADU analysis pipeline on 160 nodes of the DOE Science Grid at ANL and Pacific Northwest National Lab. The team successfully analyzed 59 DOE microbial genomes — a task that required more than 200,000 Blast jobs and generated 50 GB of data — in 11 hours. This was a significant improvement over the performance of a 32-processor cluster, which required an average of five hours to process a single prokaryotic genome, according to Maltsev.
Maltsev said the GADU team is currently modifying other bioinformatics applications, in addition to Blast, to run on the Science Grid backend, and also plans to add additional DOE computing nodes to the system. “Once you have the gateway, you can expand,” she said.
While the computational capacity of the grid should help relieve the obvious number-crunching bottleneck in high-throughput genomic analysis, Maltsev said the GADU developers also plan to exploit the storage and data distribution benefits of the grid infrastructure. GADU’s grid expert partners were especially helpful for this part of the project, Maltsev said: The National Center for Supercomputing Applications (NCSA) contributed storage space on its Starlight optical network system and the Globus team modified its Chimera data-flow middleware, which documents “data provenance” so users can assess the validity and reliability of data and computational resources.
Storage capacity and management is a key component of GADU, Maltsev said, and the system enables both permanent storage and temporary storage — an important requirement for genomic analysis, she noted, because intermediate outputs of parsers and analysis tools can be just as important to retain as sequence data and annotations, but require considerable space. Analysis of an “average” prokaryotic genome of 4,000 genes, for example, would require 1.2 GB of temporary storage and 1.0 GB of permanent storage, she said.
If the project continues as smoothly as it has so far, GADU and the grid will prove to be a perfect match, according to Maltsev. “First, you can analyze the data very quickly; second, there is storage space so you can store the data you are acquiring; and third, you can distribute the data you are acquiring — it’s a dream environment.”
Bringing Biologists on Board
Of course, there are always a few bumps on the road to realizing such dreams. “Some of the protocols are incompatible,” she noted, “but people are trying to resolve these issues. For example, we were trying to use the PNNL nodes, but we need to do some additional work to make the architectures compatible.” In addition, she said, the DOE Science Grid architecture was initially “not very accommodating for the types of data that we have, because our data is embarrassingly parallel with huge amounts of sequences” — a challenge that required a fair bit of “restructuring” of the grid architecture to resolve.
But the relatively novel technological territory on both the grid and the bioinformatics sides of the joint project helped make things interesting, according to Maltsev. “While they are observing how we do bioinformatics, the Globus group also finds improvements to what they’re doing,” she noted. And, as far as grid proponents are concerned, this symbiotic interchange is really the mission of the project: Not only do biologists need applications that will run on Globus and other systems before grid-enabled bioinformatics takes off, but the caretakers of the grid infrastructure must ensure that the underlying technology is compatible with biological data.
Based on the success of GADU so far, neither Maltsev nor Foster foresee any hurdles to bringing more bioinformatics projects onto the grid. “Everyone thinks they’re special,” Foster said, referring to new user communities adapting to grid computing, “and in some sense they’re not because the basic technology requirements are the same.” Foster did note that life science research — even in academia — does require a higher level of security than the physical sciences. With this in mind, “we’re putting a lot of effort right now into techniques that will allow communities to manage their [security] policy on a community level while at the same time allowing for more flexible local site security policies,” he said.
Maltsev said the GADU development team is on track to launch an alpha version of a public GADU server before the end of the summer, which will give users access to the Science Grid’s computational power and also enable users to assemble their own analytical pipelines. “Everybody is using Blast and Interpro and transmembrane prediction programs in various combinations, so what we will try to do is to allow them to combine the tools in the pipeline that will be of specific interest to them,” she said.
GADU, as well as other grid-enabled bioinformatics projects, may offer “a chance to change the sociology of genetic sequence analysis,” Maltsev opined, by eliminating the computational barriers that many small universities and research groups currently face. “If they were able to analyze huge amounts of data using grid technologies and a public server, they could do the analytical part of the investigation without building all these clusters and huge databases,” she said.
GADU is currently funded by the NCSA Alliance, NIH, the University of Chicago, and ANL. In addition to Maltsev, major contributors to the project include Alex Rodrigez, Veronika Nefedova, and Jens Vockler. Maltsev said she plans to seek additional funding for the project following the alpha release of the GADU server.