Intel Advances Cancer Treatment with HPC, Big Data, Open Source
Intel hopes to take an employee's painful personal healthcare journey, combine this knowledge with its advanced scale computing and partnerships with medical leaders, and transform treatment for the one million Americans who get cancer each year by leveraging big data to make personalized medicine a reality.
At the Intel Developer Conference, the company unveiled a partnership with the Knight Cancer Institute at Oregon Health & Science University (OHSU) to launch the Collaborative Cancer Cloud. This open platform-as-a-service solution is a "precision medicine analytics platform" where institutions can securely share genomic, imaging, and clinical data, Eric Dishman, director of proactive healthcare at Intel, told EnterpriseTech.
By 2020, the goal is for patients to receive personalized treatment recommendations within 24 hours – not the days or weeks it takes today to get more generalized treatment plans that, as Dishman personally learned, can be incorrect and dangerous.
Rather than moving data – and dealing with regulations such as HIPAA – and organizations' concerns over proprietary knowledge, the Collaborative Cancer Cloud "sends the analytics out to the data," Dishman said. "It protects in both ways; it protects the raw data … but it also protects the intellectual property at the site. Another cancer center can say you're an approved partner but they're not going to know the secret sauce. It's really about securing both sides of the coin."
Moving data also takes time, a resource physicians – and especially cancer patients – do not have, said Dr. Brian Druker, director of OHSU Knight Cancer Institute.
"If you just think about the cancer genome it would take 21 days to move it over our current fiber optic cables. If you scale that to 21 genomes a day, it became clear to us you needed to keep the data where it is and move the analytics to the data," he said.
Using current technologies, sequencing the genomes of every American diagnosed with cancer in 2015 would take 15,000 server hours and run up a $53 million electric bill, added Dishman.
To address the challenge, Intel designed an exacloud HPC infrastructure for OHPC's cancer institute using, in part, technologies out of its labs. The developer plans to to open source some of the tools it used, beginning in the first quarter of 2016, said Dishman.
The university already had a one-year-old Intel exacloud HPC infrastructure but the cancer cloud taxed this system. Intel added 2U nodes – each of which provides two x16 PCIe slots – to the existing 4,848 Intel Xeon processor cores and 23.8 TB of memory spread across the 202 compute nodes. One new node hosts Intel Xeon Phi coprocessor cards; the other, designed for expansion, may eventually house graphics coprocessors, according to Intel.
A hybrid network of InfiniBand and 10 Gigabit Ethernet (10 GbE) supports the exacloud – and the huge amount of traffic running into the datacenter.
Intel attached scratch storage to the front plane of each Intel server chassis. Each compute node includes a 480 GB Intel Solid State Drive 520 series with 4.6 terabytes of hard disk storage. Long-term storage – 240 TB of raw storage per 4U of space – spans the cluster via Intel Enterprise Edition for Lustre distributed file system. Lustre needs three dedicated servers, which also reside in the exacloud.
This compute power will speed up genomic sequencing, allowing physicians to spend their time devising personalized treatments for patients – not chasing down faxes, fellow doctors, or reading professional papers and magazines, Druker said. And by reviewing workflow, then improving software so it performs up to 1000-times faster, Intel expects to automate some basic processes, such as determining whether a mass is benign or potentially cancerous, to allow doctors more time to work with patients, noted Dishman.
Said Druker: "It's a huge difference in the way we do business. If I'm seeing a dozen patients in clinic that means that evening, the next day, I'm spending hours and hours of what to make of the data. If that becomes automatic, that frees me up to do much more interesting research projects on the data and it accelerates our progress against cancer and other diseases. As Eric says, it lets me do my work taking care of patients. If I'm spending my time running over to the radiologist, trying to track down a radiologist to run over films with me, how useful is that? I could have seen three more patients in that time."
The Collaborative Cancer Cloud is up and running today, and Intel expects two other cancer center partners to join by the first quarter of 2016. Eventually the center should act as a template for all diseases that have genomic causes, Druker added. Indeed, the cardiology department at OHSU already voiced interest in this solution.
"This will scale to most human diseases," he said.
Nearer term, it will incorporate the growing plethora of Internet of Things devices designed for and complementary to healthcare and personal monitoring. The exacloud HPC solution allays doctors' typical worries about data overload and this information supports researchers' treatments, said Dishman.
"Precision health is about combining them all in a digital way. The power of real-time insight from the edge combined with real-time genomic insight from the hospital and clinic," he said.
It's a personal message for Dishman, who was misdiagnosed and mistreated for more than two decades. Finally, after his genomes were sequenced in a process that took three months at a time he was not expected to live long, his doctors determined Dishman had been mistreating him for years.
"Three months later, I was cancer-free," Dishman said, after he received a kidney from an Intel colleague.