In 2003, Compugen co-founder Simchon Faigler purchased the rights to the company's Bioccelerator line of FPGA-based bioinformatics systems, creating a new firm called Biocceleration dedicated to selling and servicing the accelerated informatics system.
Two and a half years later, Biocceleration is now shifting its focus from the flagship hardware platform that has been on the market for more than a decade toward a software product that Faigler claims can offer comparable speedup to the FPGA boxes, but at a much lower cost.
The US Patent and Trademark Office a key Bioccelerator account that the company inherited from Compugen has purchased "multiple licenses" of the new software, called GenCore 6.0. Faigler said that this agreement the first for the new product validates the company's strategy as "proof that a significant customer is looking into it and understands that the shift is here, and it's time to shift from hardware accelerators to software solutions."
Faigler stressed that Biocceleration is "still focusing on accelerating genomic database searches," but added that the company is addressing a "significant shift" in the marketplace from hardware-based accelerators to optimized software that can run on general-purpose computers.
According to Faigler, this shift is based on simple economics. "Once there is a technology that runs on a $3,000 or $4,000 Linux machine that gives you similar performance to a $100,000 hardware machine, there is no economical reason to buy the hardware accelerator," he said. "We think that the technology is here."
"Once there is a technology that runs on a $3,000 or $4,000 Linux machine that gives you similar performance to a $100,000 hardware machine, there is no economical reason to buy the hardware accelerator."
GenCore 6.0 supports a number of common bioinformatics search tools, including Smith-Waterman, ProfileSearch, ProfileScan, HMMsearch, HMMScan, GeneWise, and Biocceleration's Frame+ and ProfileFrame+. According to Faigler, the software offers a 10-fold speedup over standard implementations of these algorithms by taking advantage of the SSE2 (streaming SIMD extensions 2) instruction set on Intel's Pentium chips.
By programming at the "very low level of the processor," Faigler said that Biocceleration can treat the chip as a vector processor, parallelizing the algorithms to achieve "a very significant acceleration." The company cites a "typical" benchmark on its website, in which GenCore 6 running on a Linux cluster of eight 3GHz Xeon processors completes a Smith-Waterman search against UniProt 4.2 with a protein queries of 500 amino acids in about 230 seconds (further benchmarks for GenCore 6.0 are available at http://www.biocceleration.com/GenCore6-Benchmarks.html).
Faigler acknowledged that this performance is not as fast as that achieved with an FPGA-based accelerator, but noted that the speedup is "becoming very close to the performance that we used to achieve with our hardware accelerators."
The deciding factor for most customers, he said, will be price rather than performance a fact that the company is betting on as it migrates from a business model built upon the sale of high-end hardware systems.
"If you look at the hardware accelerator market, we knew over the years that we had something like 50 to 100 potential customers that were willing to pay $100,000-plus. But [for] the software solution, we feel that there might be 1,000 or 2,000 potential commercial customers that would be willing to pay, say, $10,000 to $20,000 for a package" a potential four-fold increase in market opportunity.
"We'll continue to sell the hardware accelerators, but we keep telling our customers that we don't recommend buying them," he said. "We have the capability to support them, we have the capability to manufacture these machines, but from a price/performance point of view, and from where the market is going, we feel it's not the right choice right now."
In addition, Faigler said, the company is offering "attractive" licensing agreements for customers who want to run the software on Linux farms. For customers purchasing 64 to 128 GenCore licenses for use in a cluster, Biocceleration offers a "deep volume discount," he said.
The company is also making the software available for free to academic users.
Biocceleration is not the only bioinformatics company working at the level of the processor to speed sequence searching. Scalable Informatics currently offers an optimized version of HMMer that is up to 2.5 times faster than the standard version of the software, and the company is developing SSE-based methods to improve that performance even further.
Vipin Chaudhary, an associate professor in the department of computer science at Wayne State University, is working with Scalable Informatics on the project. Chaudhary said that while SSE-based programming can offer "substantial speedup" for some algorithms, such as Smith-Waterman, it's generally "not a good substitute" for hardware-based systems that can offer acceleration of 50 to 100 times over standard CPUs.
Chaudhary said that in his experience, a 10-fold speedup of SmithWaterman using software alone is "unlikely," but an acceleration of 5- to 8-fold is "reasonable."
He added that few firms currently offer accelerated software based on SSE programming, because it's "basically assembly programming … and very few people really want to look at the code that closely." In terms of difficulty, he said that SSE programming is still "easier than programming an FPGA," but more difficult than most bioinformatics developers would feel comfortable with.
"There's room for people to have a business based on this technology," he said, "but it may not be a very big one."
Unlike Biocceleration, Scalable Informatics still thinks there is a sizable market for hardware accelerators, but the company has also identified price as the limiting factor in the current market. Rather than move completely toward software, however, Scalable is developing a hybrid hardware/software platform that it hopes to sell for under $10,000 about the price of a high-end server, Chaudhary said.
Biocceleration's future plans may involve porting its software to run on distributed networks of desktop PCs within pharmaceutical companies. "I think the next major step for us will be to release a grid-enabled package that can run on standard grid packages that people are using today," Faigler said, citing United Devices as an example of a potential partner in this area.
While Faigler declined to provide a specific timeline for the launch of such a product, he said it's possible that it could be available before the end of the year.
Bernadette Toner ([email protected])