By Jennifer Patterson Lorenzetti
January 13, 2003
| What if the efficacy of a drug for a particular patient could be predicted simply by looking at that patient's genome sequence? Or what if drugs could be tailored to attack diseases with a genetic component?
Considering that most diseases have both genetic and environmental causes, these are vital questions — questions that Mountain View, Calif.-based Perlegen Sciences Inc., an offshoot of DNA microarray pioneer Affymetrix Inc., tackles on a daily basis.
Standing on the shoulders of the Human Genome Project, Perlegen has made important advances in resequencing DNA and identifying SNPs (single nucleotide polymorphisms), to the point that an entire human genome can be deduced in about 10 days. These genetic scans are obtained from scanning whole wafers of Affymetrix Inc.'s DNA microarrays, rather than merely individual chips.
In the past year, Perlegen has amassed a library of genome scans of 25 individuals. They have focused on some 1.7 million SNPs (out of a total of more than 3 billion bases) in each genome, a powerful basis on which to compare genetic differences across populations in the hunt for disease genes. It is also an extraordinary volume of data to store.
Making this storage process run smoothly is the job of Greg Brandeau, chief information officer of Perlegen. Brandeau is no stranger to managing vast amounts of data, having previously spent five years as vice president of computer operations for Pixar Animation Studios, best known for its "Toy Story" movies. But this is something bigger still.
According to Brandeau, Perlegen creates a staggering 2.5 terabytes of scanned data each week. Because the process of scanning each wafer to record the location and intensity of millions upon millions of discrete signals is destructive, keeping the results of each experiment is of critical importance.
"We couldn't re-create it without spending lots of money," Brandeau says. So he has constructed a three-part storage system that ensures that images are captured, available for work, and stored safely for the future.
The first component is the NAS (network- attached storage) database. "Most people think this is crazy, but it works fine for us," says Brandeau, who has used the power of ethernet switches to buck the conventional wisdom that NAS would operate more slowly than disk drives attached directly to a computer. This storage component holds 4TB and consists of two F880c machines from Network Appliance Inc.
It's not a toy: Having DNA microarrays available on wafers (as displayed by Greg Brandeau, Perlegen's CIO) has enabled Perlegen to scale up genotyping analysis.
In fact, Brandeau has found that "we can take a snapshot of our 4TB database in less than a minute." The database keeps track of the location of each scanned image, as well as tracking how far along the data analysis pipeline that particular image is. In this way, control of image processing and storage tracking happens automatically.
Next is image storage, a process Brandeau describes as "scratch space from the scanner." At Perlegen, images reside in the image storage system until they have worked their way through the data processing pipeline, after which they can be removed to make room for new data. Image storage is handled by a bank of four Network Appliance F840c machines.
And the importance of reliable data storage is well understood by Network Appliance co-founder Dave Hitz. "Reliable data storage is important in every industry, but especially so in biotech," he says. "Losing the wrong data at the wrong time ... could jeopardize an entire research project. In this environment, nothing is more important than data safety."
At the same time that images are written temporarily to image storage, they are written permanently to Perlegen's hierarchical storage management (HSM) system and tape jukebox. This permanent storage area consists of a Sun E450 server, running Sun SAM-FS HSM software, connected to an ADIC Scalar 10K LTO jukebox, which can hold up to 340TB. Substantial storage capacity is critical, as last year alone, Perlegen created an estimated 110TB to 120TB of data.
Ultimately, three tapes are created from this system — one that resides in the jukebox itself, one offsite, and one that is stored. The system is linked by a Cisco Catalyst 6513 and a Sun E6500 Oracle database server. Finally, a compute farm — the engine that performs data analysis — is linked to the entire storage system. Perlegen operates in a Windows environment running .NET, and Brandeau says this choice is very cost effective.
Brandeau feels lucky that Perlegen started up when it did in October 2000, especially from a technology standpoint. For one thing, many of the components that he needed to construct the IT system were available commercially, including gigabyte ethernet and large-capacity drives. "[We took] very good off-the-shelf equipment and made it work for us," he says. And DNA micro-arrays were available at the wafer level, rather than the chip level, enabling Perlegen to make rapid progress in scanning DNA variations.
|Analyzing the Genome
|Although the storage of images is a critical undertaking for Perlegen, this is far from its primary mission.
If he were starting such an endeavor today, Brandeau says he would change little, though he concedes he might rely more on disk than on tape. "Disk drives are getting denser and faster rapidly," he says. In Brandeau's estimation, disk speed is growing four times as fast every 18 months compared with twice as fast for tape. Because of this, Brandeau says he "might just buy enough spinning disk" for storage, using tape only for backup.
This is quite a change for Brandeau, who joined Perlegen in early 2001. The challenges of recording the millions of DNA sequence variations that govern human disease susceptibility are a far cry from the work he did at Pixar helping set up the computer system that would be used to create the digital characters of "Toy Story" and "Monsters Inc."
"The learning curve is very steep," he acknowledges, explaining that, at first, he used to think that he was beginning to understand the complexities of his new work, only to later learn that he had just begun to master the buzzwords. Fortunately, he has daily access to Perlegen's chief scientific officer, David Cox, former director of the Stanford University Genome Center, whom Brandeau praises as "an excellent teacher" helping to bring him up to speed in the transition from pixels to base pairs.
Brandeau is excited by the work that Perlegen is doing, likening it to other pivotal points in the history of science. "Before Galileo had the telescope, no one could see the motion of the planets, but then [once he did, we] could discover the laws of physics," he says. Likewise, he compares Perlegen's work to having "a super high-powered microscope that can look at DNA," paving the way for pharmaceuticals that are tailored to specific conditions and specific populations.
"We are only at the very beginning of the genetic revolution," he says.
Monstrous: Perlegen's IT network handles 2.5 terabytes of scanned data per week.
Jennifer Patterson Lorenzetti is a freelance science/technology writer and owner of Hilltop Communications in Centerville, Ohio. She can be reached at firstname.lastname@example.org.
PHOTOGRAPH BY ANNE HAMERSKY