Researchers at Texas A&M University’s animal genomics group really needed a Blast farm. “We had a community of people complaining about not being able to do Blast searches,” said David Adelson, an associate professor in the department. The 30 or so researchers working in the group had been running thousands of Blast searches at a time through the NCBI site, but when NCBI lowered the limit on the number of searches it allowed per day, they were left in the lurch.
Although the research group had access to a supercomputer, “It was all command line, and required mastering Unix and learning shell scripting. We had to train people to get it to work,” Adelson said.
So after considering their available options — and their limited budget — Adelson and his colleagues settled on the idea of a cluster, but wanted to add a web-based, point-and-click interface so that non-experts could handle the load management and system administration aspects of the system. In addition, Adelson said, he was toying with the idea of trying out Apple’s new Xserves. “I was familiar with AG blast and was thinking, ‘Wouldn’t it be nice if we could get an Xserve cluster?’”
After a presentation by Apple convinced him that an Xserve cluster offered “unique aspects for Blast searching,” Adelson spoke to consulting firm BioTeam to deliver what he described as a “full wish list” — a web interface on top of the load management layer on top of the cluster. After further discussions and some brainstorming, they arrived at a low-cost system that provided everything Adelson had requested, along with some additional features.
BioTeam first ported Sun Grid Engine to OS X, and then integrated the open source PISE (Pasteur Institute Software Environment) interface with the Xserve cluster to give Adelson and his colleagues a Blast farm that doubles as a bioinformatics portal. Researchers have access to the full EMBOSS suite as well as a number Altivec-optimized tools, including Hmmer, Fasta, and Clustal. Jobs are submitted via the web to PISE, which then sends them to SGE, where they are split among the cluster’s 10 CPUs.
Installation of the system took less than a month spanning December through January, Adelson said, and the system is just about ready to enter production mode.
Acknowledging that he could have built a similar system on his own using the same set of public domain tools, Adelson noted that the time saved by hiring an outside party to install the cluster was worth the additional expense. “I take the view that time is money,” he said. “We had an urgent need for this system, and I would still be bumbling around getting the cluster configured if I did it myself.”
While Adelson is pleased with the current system, he still has a few remaining items to check off his wish list. “I would like to find the equivalent system for microarrays,” he said.