Skip to main content

MPT Eyes Life Sciences for Initial Launch of Parallel Processing Tech; Seeks Beta Testers

Premium

Massively Parallel Technologies, a DARPA-funded high-performance computing startup, has identified bioinformatics as the initial commercial market for its number-crunching capabilities, but it is first looking for some willing participants to test-drive its technology.

MPT is enlisting beta customers for its first product, an implementation of NCBI's Blast that runs on its in-house Virtual Power Center (VPC) and is accessible via a web-based interface. Kevin Howard, MPT's president and CTO, told BioInform that the company is first targeting academic researchers for the online system, and will eventually market a version of its technology for biotech and pharmaceutical firms who prefer to do their analysis on their own clusters.


"For the price of a cell phone, you can get more work done than if you owned your own machine."

Howard said that the company hasn't set a target date for when it plans to launch Blast VPC commercially, but it doesn't appear to be in any rush. Funded primarily through four grants totaling $1 million from the US Defense Advanced Research Projects Agency, MPT has already taken six years to bring the technology to this point, and is taking a similarly cautious approach to readying it for commercialization.

"We're going through an extensive beta test period," Howard said, adding that MPT is seeking additional beta customers "so that we can make sure that our stuff is not only rock solid, but that we know the level of our user base and can really get our act together."

Howard said that the key to MPT's approach is a "new mathematical simulation of how communication works in parallel processing." The software-based technology is able to improve cross-communication between processors for any kind of off-the-shelf hardware, and the company claims that it enables efficiency improvements by multiple orders of magnitude over traditional parallel processing technologies.

MPT's Blast Performance Comparison

Machine: 1,466 MHz AMD Athlon (127 processors)

Run parameters: 55,344 input query sequences (10,626,048 total letters) searched against NCBI's UniGene database (45,717 sequences; 81,550,205 total letters)

NCBI-TBlastN
WU-TBlastN
MPT VPC
(running NCBI TBlastN)
14.8 days
11.7 days
3.5 hours

For Blast, which Howard described as "the low-hanging fruit" in the bioinformatics market, the company's technology enables "linear speedup" up to more than 250 nodes. MPT also plans to run the Fasta, HMMer, and Smith-Waterman algorithms on its cluster when it launches Blast VPC later this year. In addition, MPT has targeted molecular dynamics algorithms like Amber for launch in mid-2006, with plans to support high-throughput docking algorithms around six months after that.

MPT has opted for a hosted services model because its initial target market "hates computers," Howard said. The company hopes to attract business from small labs and academic groups that don't have the resources — or desire — to install their own cluster for Blast searches. While pricing for the commercial service has not yet been finalized, Howard estimated that it would run a typical lab around $100 per month.

"For the price of a cell phone, you can get more work done than if you owned your own machine," he said.

While the bioinformatics ASP model hasn't proven successful in the past, at least one of MPT's beta customers thinks the company's business model makes sense. Alecksandr Kutchma, a bioinformatics research assistant professor at the Barbara Davis Center for Diabetes Research at the university of Colorado Health Sciences Center, said that his lab would be willing to pay "at least that much" for access to the service.

Kutchma said that his group currently runs all its Blast searches on a single Pentium 4 machine, which it is quickly outgrowing. Some of the lab's large queries can take up to a month to complete — a timeline that is "really not feasible," he said.

Kutchma and his colleagues looked into buying a dedicated cluster, and also considered renting time on other supercomputers, but found that the MPT system offered an attractive alternative. In a comparison between Blast VPC, NCBI's online system, and the UCHSC Blast server, "we noticed that as the queries got larger and larger, MPT's Blast became much faster compared to NCBI and our system."

Howard said that the most likely customers for MPT's service will be labs with a query volume that just surpasses the limit of NCBI's online capabilities. Faced with the option of buying and maintaining a cluster, downloading the software, updating the databases, and writing scripts to manipulate the data, these labs soon find that the costs can quickly add up to hundreds of thousands of dollars — even though NCBI's software is free. "It may be free to use, but is it really free if you have a six-month grant and it takes 12 months to run your compute job, or if you have to buy $100,000 in computing equipment, or pay a grad student to maintain your system?" Howard asked.

Kutchma acknowledged that "casual users" who just run a few Blast queries a week may prefer NCBI's interface to MPT's service, but said that that the company has been "perfectly willing to cater to the specific requirements of advanced users" like his group, which runs up to hundreds of thousands of query sequences at a time.

— Bernadette Toner ([email protected])

Filed under

The Scan

UK Funds to Stay Ahead of Variants

The UK has announced a further £29.3 million to stay on top of SARS-CoV-2 variants, the Guardian reports.

Push for Access

In a letter, researchers in India seek easier access to COVID-19 data, Science reports.

Not as Cold

Late-stage trial results are expected soon for an RNA-based vaccine that could help meet global demand as it does not require very cold storage, the New York Times writes.

Genome Research Papers on Microbes' Effects on Host Transfer RNA, Honeybee Evolution, Single-Cell Histones

In Genome Research this week: influence of microbes on transfer RNA patterns, evolutionary relationships of honeybees, and more.