Adaptive Genomics, a startup founded by Virginia Tech alumni, has thrown down the gauntlet in the quickly crowding market for FPGA-based bioinformatics tools with claims that it has set a new record for aligning human chromosomes X and Y with the Smith-Waterman algorithm.
The company said this week that its base 2U rack HyperSeq system, based on field programmable gate array technology, aligned the two chromosomes with the computationally intensive Smith-Waterman algorithm in under 30 hours — a 584-fold improvement over the same search on a 64-bit AMD Opteron-based PC, which would take two years, according to the company.
Indeed, it appears that the firm has surpassed other benchmarks using the same alignment. Starbridge Systems, another developer of FPGA-based systems, claimed in 2004 that its Hypercomputer took five days to align chromosomes X and Y [Bioinform 04-12-04].
Scott Harper, lead researcher and hardware engineer for Adaptive Genomics, said that the company’s system outperforms competing FPGA-based systems “by a factor of 10,” though he acknowledged that the lack of standard benchmarks in the industry makes it difficult to compare across platforms.
“We can compare it to a general CPU we have here or a cluster that’s reported general numbers, but if there were an industry standard, we’d love to run that benchmark,” he said.
The company does not have a shortage of competition: Last month, CLC Bio and Mitrionics announced plans to launch FPGA-enabled bioinformatics hardware systems [BioInform 06-02-06]; in May, Singapore-based startup Progeniq launched BioBoost, an FPGA-based system that it claims offers a 10- to 50-fold speedup over conventional processors; and late last year, Cray announced its intention to target the bioinformatics market with FPGA technology [BioInform 11-14-05].
All these newcomers face stiff competition from TimeLogic, an early mover in the FPGA-based bioinformatics market with an established customer base for its DeCypher accelerator, as well as a relatively new offering called CodeQuest, a desktop workstation that includes a DeCypher accelerator card. [BioInform 06-27-05].
Adaptive Genomics intends to establish an edge in the market with improved performance and shorter cycle times between new releases. “We want to get new stuff out quicker,” Harper said. “We’re not going to sell you something that works today, but two years from now somebody’s going to sell you something better. We’re going to sell you something that works today, and two years from now we’ll let you update to something that works better with the same piece of hardware.”
“We can compare it to a general CPU we have here or a cluster that’s reported general numbers, but if there were an industry standard, we’d love to run that benchmark.”
Harper said that Adaptive Genomics plans to offer customers a service contract “that would allow you to keep up to date with regular maintenance releases. As we do new algorithms, we’re proposing selling those as separate modules.” The HyerSeq is linearly scalable “to an extent that we haven’t reached yet,” he said, “So the upgrade path we’re looking at would probably be to buy more modules and add them to your existing system.”
Harper said that the company next plans to adapt hidden Markov models for the system and is especially interested in implementing “custom algorithms” for customers developing new bioinformatics methods. “A lot of the stuff that people do in bioinformatics is matrix based,” he said. HyperSeq “is really well suited to that sort of application, and anything that uses that kind of algorithm would probably run pretty well on our system, and we’d like to port some more algorithms to the system and see how well they do,” he said.
Adaptive Genomics has several beta customers for the HyperSeq but has not yet started selling the system. Harper said the company should start shipping the system in three months.
Martin Gollery, associate director at the Center for Bioinformatics at the University of Nevada at Reno who formerly served as director of research at TimeLogic, said that while Adaptive Genomics’ reported time for Smith-Waterman does appear to be around 10 times faster than competing systems, most users don’t run Smith-Waterman.
“I would be curious to see what they can do with BLAST, ClustalW, HMMs, etc.,” he wrote in an e-mail to BioInform.
Gollery said that the recent spate of development in the FPGA-based bioinformatics market should help drive innovation in the field. “I am hoping the competition makes everyone’s products better, and we can get away from these energy-hogging clusters,” he said.