NEW YORK (GenomeWeb) – Dell announced this week that it has upgraded its hardware platform specifically designed for genome analysis, giving it the power to process around four times as many genomes per day as the first version. It has also cut the time taken to analyze samples at 10x coverage by 40 percent, down to 11.5 hours.
The High Performance Computing System for Genomic Data Analysis is an update to Dell's Genomic Data Analysis Platform, a preconfigured hardware platform developed with the company's partners in the genomics space.
"This is about taking genomics to the front lines of patient care," Charlotte Hovet, Dell Healthcare's Medical Director, told GenomeWeb. "The vision is that someday every hospital, every lab should have the technology to handle genome sequencing and analysis."
The idea is to provide the hardware for the organizations looking to take sequencing from a research-use-only tool into routine clinical use. A big part of that is speed. Dell has put in faster Intel Haswell processors and better, faster memory storage. "The same analysis pipelines that our customers ran on the first iteration run faster on the second," Glen Otero of Dell Research Computing said.
But the platform is also meant to take the complexity out of building a system capable of handling high-throughput genomic analysis. "With Dell there's [an] infinity of options," Otero said. "What we've done with this solution is taken the best practices of what we learned working with our customers," including the Translational Genomics Research Institute (TGen), Arizona State University, and the University of Cambridge. It's highly customizable, he said, but it's also standardized and ready to ship.
Customers can order from four to 40 nodes, with pricing based on the number of nodes. Dell did not provide an estimate for the price per node.
That standardization could be an advantage to carving out share in the clinical sequencing hardware market, TGen VP of Technology James Lowey told GenomeWeb. "It's easier to do things like validation when you have a consistent platform you're working on. Part of the goal of a system like this is to democratize it and give it to lots of people. When you buy a whole system under a single part number, when you want to replicate that, it makes it that much easier to do.
"If you're a big healthcare system and you want to implement genomic testing in your own CLIA-certified labs, it would make it that much easier to purchase one thing without having to validate it every time you put a new one in. Anything to make that process easier is a good thing," he said.
TGen has been the major proving ground for Dell's genomics-focused hardware and Lowey said the institute is already in the process of upgrading to the new system. Dell began working on the GDAP in 2009. "The goal was to produce a machine to give results in a clinically relevant timeframe," Seth Feder, Dell director of research for health sciences, said. "In the research field it's acceptable to wait a week for sequencing. As this thing moves into the standard of care over the next five to 10 years, we need to get [time to results] down to hours."
What Dell came up with was a "supercomputer in a rack" about the size of a refrigerator. "It's a Linux cluster and it comes integrated with all the right parts, picked specifically for genome analysis workloads," Otero said. "We've taken a lot of Dell's components on the server side to get as many cores as possible inside the container." He added that the hardware is integrated with cluster manager software to manage, install, and upgrade the software on the node and help deploy and install applications.
Essentially, Dell took care of the hard work of putting best practices into a finished product. "No one else has done that," Otero said. The idea is to get away from putting a catalog of parts in front of a customer, which may not have the in-house expertise to choose wisely among the possible combinations of components. Otero said he's done that kind of work for almost 20 years and insisted that Dell come up with a suite of features that would work right away for genomics customers.
That ready-to-use design framework hasn't changed with the new system, but the components have, he said.
One of the improvements has been the inclusion of Intel Haswell E7-8860 v3 processors. "We've picked a clock speed that is appropriate for the genomics space," Otero said. "You don't necessarily need the fastest, highest-performing, power-sucking processor," although customers can include higher performance processors if they choose.
Lower energy usage equals savings for places like TGen. "In our case we'll probably be able to save over $100,000 per year in energy bills," Lowey, TGen's VP of technology, said. "We have a system today that consumes pretty close to $300,000 per year in electric al costs and we're looking to cut that down [to around $200,000]."
Dell has calculated the wattage required to analyze each genome on both versions of its platform. The energy required to analyze each genome at 10x coverage on the first version of hardware was 7.42 kilowatt-hours per genome; for the new version Dell has reduced it to 2 kWh per genome.
Improvements to the system's memory will also give researchers a choice when it comes to assembling genomes. The first iteration was only designed for short read alignment, Otero said. "Now with more memory per node, genome assembly becomes a possibility."
But the ultimate measurement is about increasing the number of genomes per day in the most efficient manner, said David Bump, product manager for the HPC System for Genomic Data Analysis.
The new platform can process 78 samples for whole-genome sequencing at 10x coverage in 11.5 hours, yielding a total of 163 genomes per day. Dell also evaluated the performance at 50x coverage, which increases the time to 34.3 hours and decreases the number of genomes per day to 54. The first version took 19.5 hours to process 30 samples for whole-genome sequencing at 10x coverage.
That's the kind of time difference that could have clinical impact, Hovet, the Dell Healthcare Medical Director, said. Before joining Dell, Hovet worked for 20 years in family practice. She described how neuroblastoma and medulloblastoma, both pediatric cancers, were considered incurable. Supporting a TGen pediatric precision oncology program, Dell's hardware has given patients a chance to get their tumors quickly sequenced and appropriately treated. Now, almost 60 percent of patients are going into remission, Hovet said. As with these and many other diseases, "the faster the better" for getting sequencing done.
TGen expects the new hardware will accelerate analysis by anywhere from 10 to 30 percent, right out of the box. "When we start doing some optimizations we expect to see even greater gains in reducing time for analysis," Lowey said.
And while Dell is on the second iteration of its genomics-focused hardware, most of its competition hasn't yet turned its attention to the space.
"All the major players have offerings," TGen's Lowey said, "but nothing as comprehensive and cohesive as what Dell offers." IBM, HP, Lenovo, and Cisco are all supporting customers in the genomic space, he said, but the Dell systems offer what he considers unparalleled longevity. "We generally run systems in mainline production for a few years then run them through development for another two years after that. We're going to get a least five years of life out of the systems."
He added that he thinks x86 architecture will be the driving force in genomics, even as Intel and others develop specialized processors preloaded with algorithms for genome analysis. "The key difference is flexibility," he said. "When you buy [an Edico Genome] Dragen or [other] FPGA solution you're locked into algorithms that have been programmed into the hardware, unless you have access to people who know how to code it."
Dell is already taking improvements from its high-performance computing lab and testing how genomic analysis pipelines might run better with them. File systems, processors, accelerator cards, and flash memory are all being evaluated.
Moreover, Dell is looking at putting genomics-focused hardware in a desktop configuration.
"What we're talking about here is a rack," Feder said of the HPC System for Genomic Analysis. "Not every customer needs or wants to buy that." He said Dell is already showing off a desktop workstation at Supercomputing 15 that has enough power to do whole-genome sequencing.
"It's not just going bigger and faster, but shrinking its form factor so it could sit in a clinic or lab on a desktop or under a desk," he said.