Mercury Computer Systems is partnering with researchers at Boston University to accelerate computational drug-design algorithms using the Cell Broadband Engine processor.
The Cell BE, developed by Sony, Toshiba, and IBM to power the Playstation 3, includes a central processing core and eight co-processors that work together to speed up video games as well as other compute-intensive applications. Through a partnership with IBM, Mercury is marketing the chip for high-performance computing in a number of markets, including life sciences.
Mercury already has a strong customer base in medical imaging, and has recently created a new biotech group with the goal of moving into drug discovery. Mirza Cifric, director of the biotech group, described the nascent effort as an “internal venture” for Mercury.
“It’s something we find promising and hope that it will grow to a significant size,” he said. “It’s not a developed market, so there really isn’t a good comparison to our other vertical businesses.”
Nevertheless, the company views its collaboration with Boston University as an important proof of principle for the potential of the Cell BE processor. Mercury worked with BU’s Structural Bioinformatics Lab, led by Sandor Vajda, to port an application called fragment-based drug design, or FBDD, to the Cell.
FBDD, which maps protein binding sites in order to better predict how small molecules bind, originally ran on a Linux cluster in the lab and then moved to BU’s Blue Gene system.
After porting to the Cell, the FBDD simulation took only three minutes to run — around 10 times faster than Blue Gene in a chip-to-chip comparison, Mercury said.
Vajda told BioInform that the collaboration with Mercury “started by chance” through another collaboration Mercury had with scientists at BU. He said that he’s pleased with the results of the collaboration so far and plans to further develop FBDD so that it takes better advantage of the Cell architecture.
In addition, Vajda said, the BU lab plans to port another algorithm, a protein-protein docking program called PIPER, to the Cell processor.
David Lancia, a research associate at BU, said that porting FBDD to the Cell was relatively difficult, and required the help of Mercury engineers. “The Cell is a powerful chip, but it has a novel architecture so it takes a lot of effort to structure data in a way that works well,” he said.
In addition, he noted that the Cell architecture is not suited for all bioinformatics programs and said that any future work with the processor would be limited to “algorithms that are highly computational.”
Mercury’s Cifric echoed Lancia’s comment. FBDD is “computationally extremely hungry — it is extremely dependent on computation,” he said. “The nature of the mathematics that are employed in the application happen to port very well to the Cell processor.”
While most IT firms opt for Blast or Smith-Waterman as their first proof-of-principle for the bioinformatics market, Cifric said that sequence-alignment algorithms are already “well-suited for general-purpose processors” and therefore wouldn’t see the “tremendous advantage” that FBDD demonstrated when it was ported to the Cell.
Mercury’s goal in the biotech space, he said, is to not only improve the speed of applications, but to “enable things that Dr. Vajda and others simply did not conceive originally as being possible.”
Cifric added that there were also commercial considerations behind Mercury’s decision to look beyond Blast as its bioinformatics poster child.
“There’s very little interest in the market potential for those applications,” he said. “Clearly we wouldn’t want to spend all of our time and effort on something that’s not exciting and won’t be widely adopted and interesting to the pharmaceuticals and biotechs.”
Cifric noted that genomic and proteomic applications are “too far from the market, and too far from the drug-discovery process” to attract the interest of Mercury’s target customer base in biotech and pharma.
“Clearly we wouldn’t want to spend all of our time and effort on something that’s not exciting and won’t be widely adopted and interesting to the pharmaceuticals and biotechs.”
“You do a sequence alignment of something and that’s great and that’s very valuable information, but you don’t really know until 10 years after the drug is on the market whether that particular technology or application had any influence on the ultimate outcome,” he said.
Cifric said Mercury has identified a “shift” in the industry, “one where computational efforts to accelerate drug design are coming back into the cycle of interest.” However, this renewed interest will be limited to applications “with the ability to influence discovery in the late pre-clinical stages of work, where the largest opportunity for improvement exists — in the lead generation and lead optimization stages,” he said.
Cifric declined to provide specific details regarding Mercury’s plans to commercialize FBDD or other bioinformatics algorithms, but said that the application “presents an opportunity” for Mercury to explore ways of targeting the biotech and pharmaceutical market.
He added that Mercury, which was founded in 1981 to sell specialized computer boards, has been evolving toward more of a services model in recent years.
Cifric noted that life science informatics is a “tough business,” and that few companies have built successful businesses in the market.
“We see that the opportunities of providing scientific information as a service are probably a more likely way of capturing the value in this business, rather than selling software or hardware as tools,” he said.