Fujitsu may be better known in the US for its plasma TV screens than for its bioinformatics solutions, but a small offshoot of the company is hoping to change that perception. The BioIT division of Fujitsu America set up shop just a few months ago in the Boston area with the goal of building a sales and marketing strategy for the company’s life science products in the US.
In Japan, Fujitsu has been closely involved in the life sciences for more than 20 years, said Michael McManus, vice president of the US BioIT group. This effort, which began as a reseller agreement for MDL’s software in 1980, has now grown to encompass a team of more than 200 life science informatics software engineers and sales staff in Japan. The Japanese BioIT group offers a selection of its own software products, along with a range of hardware solutions and third-party tools that it distributes in the Japanese market. In addition, Fujitsu acquired Oxford Molecular Group’s Cache molecular modeling software product line in 2000, which it has continued to distribute under the same name. Fujitsu lays claim to 35 percent of the Japanese life science informatics market, which it pegs at $3.6 billion.
Now, McManus said, the company is looking to expand this life science business into the US market, where it faces a number of key challenges — including deeply entrenched competition and very different purchasing patterns. In Japan, McManus said, customers prefer buying from a single source, so the company’s mix of hardware and software has proved successful. While this strategy will set it apart from its rivals in the US market, giving potential customers a wide range of solution choices, it also presents a unique set of obstacles. After all, Fujitsu will face competition not only from the IBMs and HPs of the life science computing world, but from the Accelryses and Lions as well.
But McManus said the company is carefully preparing for this two-front battle. One trick Fujitsu has up its sleeve is the “BioServer” massively parallel server that it engineered specifically for genomics-based drug design. The company sees the BioServer as a good way to bridge the software and hardware sides of its business, by optimizing Cache’s software, public bioinformatics applications, and other life science software that it now markets in Japan to run on the system.
BioServer packs 1,920 processors into a single six-foot rack with 2.5 Tflops of computing power. It is built with Fujitsu’s FR-V series of embedded processors, which consume only one-twentieth the power of a Pentium and one-fiftieth the power of a Xeon, McManus said. Lower power consumption means less heat, offering substantial cost savings in power and cooling. In addition, McManus said, the chips can be stacked closer together than Intel or other chips that generate more heat, saving space. The tradeoff, however, is that the FR-V chips offer only about one-fifth the compute power of a typical Intel chip.
One way Fujitsu makes up for the performance loss is by eliminating interprocessor communication. The system is managed by one of Fujitsu’s PrimePower servers, which splits up compute jobs, sends them out to each of the 1,920 processors in the system, and then collects them when the job is done. In addition, Fujitsu plans to tune the algorithms that it ports to the BioServer. The company has already modified the Gromacs molecular dynamics algorithm to run on the Bioserver, and McManus said the company has tuned the program to run three times faster than it does on an Intel chip. Fujitsu has also written versions of Blast, Hmmer, and other common bioinformatics applications to run on the BioServer, and plans to port Cache’s ActiveSite high-throughput docking algorithm to the system as well.
So far, the BioServer is still classified as a research computer in Japan, and two validation systems are installed at the company’s research labs. ZoeGene, a genomics tool subsidiary of Mitsubishi Chemical, is working with Fujitsu to validate the system. McManus said that the company does not yet have a timeline in place for introducing the system in the US, but he said the company is currently in “exploratory discussions” with several US organizations interested in using it. “We’ve had a tremendous amount of interest from people saying they pay a fortune for cooling and power for a 2,000-CPU cluster, and they couldn’t believe that this could sit in a little office,” he said. He added that the company plans to price the system at “around one-fifth” the price of a comparably equipped Linux cluster, but was unable to provide more specific pricing information.
The Boston Team
McManus was quick to admit that an integrated BioServer/Cache offering “is the picture of where we’re going, but we’re not there yet.” His task over the next few years is to build up a sales and marketing team in the US, “and to figure out if there is a market for these things in the US, and, if so, how do we best introduce them?”
So far, the Boston team comprises four people — McManus, who was previously at AnVil, along with informatics marketing veterans Jim Sullivan (formerly at Molecular Mining and Tripos), Ian Welsford (formerly at Proteome/Incyte, Molecular Mining, and Ardais), and Josh Nathan (formerly at Proteome/Incyte and Ardais). Counting the Cache development team in Beaverton, Ore., the US BioIT group currently numbers around 20 people. McManus said he plans to fill a few more sales positions in the Boston office, and expects the whole team to be in place within the next few weeks.
Nathan noted that the small group has all the advantages of a start-up environment, along with the security of being part of a global enterprise — a welcome atmosphere for the staff, which has seen its share of upheaval in the bioinformatics sector. “We have the support from the corporation, we’re well funded, but we also have the agility almost of a 20-person or less startup mentality,” he said.
McManus said that the group has about two or three years to prove its effectiveness to Fujitsu Japan. He said that he’s “hopeful” that his team will be able to generate a solid business in the US, “but I’m a pragmatist,” he added. “We have to build market awareness first.” Specifically, he said, it’s a given that the potential market for these tools is “potentially very large, but it’s the available market that we really have to decipher.”