HPCwire: The fact that Precision Medicine is the opening panel at SC strongly suggests the growing importance of HPC in making PM and basic life science research possible. Recognizing SC is primarily a technology conference, could you frame the goals of this panel?
Steve Conway: Precision medicine, also called personalized medicine, promises to transform medical practice and healthcare spending by enabling called personalized diagnoses and treatment plans that are custom-tuned for each patient’s physiology, symptoms, medical history, DNA and even lifestyle. What constitutes a good outcome for a broken hand may be different for an office worker and a concert violinist. HPC is already playing a key role in early precision medicine initiatives around the world, by speeding up genome sequencing and by making it possible to quickly sift through millions of archived patient records to identify treatments that have had the best success rates for patients closely resembling the patient under investigation. Biology is fast becoming a digital science and healthcare analytics is one of the fastest-growing new market segments for HPC. Precision medicine is happening at the intersection of biology, medical practice, healthcare economics, and data science. The expert panel at SC’16 will explore this emerging domain from these varied perspectives, with special emphasis on the major role HPC has already started to play.
This is a pretty august group:
- Mitchell Cohen, Director of Surgery, Denver Health Medical Center; Professor, University of Colorado School of Medicine.
- Martha Head, Senior Director, The Noldor; Acting Head, Insights from Data at GlaxoSmithKline Pharmaceuticals
- Warren Kibbe, Director, Center for Biomedical Informatics and Information Technology (CBIIT); Chief Information Officer; Acting Deputy Director; National Cancer Institute (NCI)
- Dimitri Kusnezov, Chief Scientist & Senior Advisor to the Secretary, U.S. Department of Energy, National Nuclear Security Administration
- Steve Scott, Chief Technology Officer, Cray Inc.
HPCwire: Today much what constitutes PM is big data analytics. Within this context: a) what are the key technologies (compute/architectures, storage, informatics, etc) being used, b) what are the big technology challenges/bottlenecks, and c) where do you expect near-term progress?
Conway: We’ll hear more about this from the experts on the panel, but in general the computer technologies being used today to support precision medicine vary from purpose-built supercomputers such as IBM Watson with its advanced natural language capability to Linux clusters with the usual processors and software. One big challenge is getting access to detailed data on large enough patient populations—some big healthcare companies are investing a lot of money today to acquire more data. Another challenge is speed. An important decision-support goal over time is for the computer to spit out efficacy curves for treatment options in near-real time, while the patient is still sitting across from the doctor. Yet another challenge is the state of the data science—there’s a big need for tools that help users understand the data better, including benchmarks to verify that the results are useful.
HPCwire: How significant is the relative lack of HPC expertise and general computational literacy of most clinical physicians and even life scientists generally? The command line is hardly a friendly place for them. What, if anything, should be done to support them and to raise their computational skill level?
Conway: One of the biggest barriers across all of HPC is the C. P. Snow “two cultures” problem, where in the case of HPC you have computer scientists and domain scientists trying to communicate with each other using different languages. In precision medicine you might have HPC vendors talking about integer or floating point operations per second, while the buyers and users want to hear about cancer detections per second. My own opinion is that in precision medicine, to be successful HPC vendors will need to bend more toward the users than the other way around. I don’t think vendors can expect users to make a big effort to become more proficient in HPC. It will be interesting to hear what the panelists at SC’16 have to say about this.
HPCwire: How should we expect delivery of PM technology to evolve? IBM Watson has received a lot of attention using a cloud-like model while many institutions have on-premise resources. How will the PM delivery ecosystem (HPC infrastructure) evolve?
Conway: Again, you’ll get a fuller discussion of this during the SC panel session, but it seems clear that an effective precision medicine environment will involve both on-premise and cloud resources, presumably integrated in a way that’s transparent to users. You’ll need on-premise resources for brute force computing and cloud resources for things including data research, records transfer and general communication. Most healthcare systems already rely on private clouds for communication among providers and between providers and patients. The brute force computing will be needed for near-real time diagnosis and treatment planning.
HPCwire: What are the two or three examples of the most advanced HPC-based PM systems used today and what makes them distinct?
Conway: Let’s start with IBM Watson. In 2011, Watson stunned a huge American television audience by defeating two human past champions of the Jeopardy! game show in a competition match. The great achievement of this digital brain was its ability to “understand” natural language — specifically, natural language expressed in the interrogatory syntax of the game show. On the heels of this triumph, IBM announced in January 2014 that it would invest $1 billion to advance Watson’s decision-making abilities for major commercial markets, including healthcare. Not much later, in May 2015, IBM said 14 U.S. cancer treatment centers had signed on to receive personalized treatment plans selected by a Watson supercomputer. Watson has contracted since Jeopardy! days “from the size of a master bedroom to three stacked pizza boxes.” Watson will parse the DNA of each patient’s cancer and recommend what it considers the optimal medical treatment, so it’s a powerful decision-support tool for healthcare providers.
The Center for Pediatric Genomic Medicine at Children’s Mercy Hospital, Kansas City, Missouri, has been using supercomputer power to help save the lives of critically ill children. In 2010, the center’s work was named one of Time magazine’s top 10 medical breakthroughs. Roughly 4,100 genetic diseases affect humans, and these are the main causes of infant deaths. But identifying which genetic disease is affecting a critically ill child isn’t easy. For one infant suffering from liver failure, the center used 25 hours of supercomputer time to analyze 120 billion nucleotide sequences and narrowed the problem down to two genetic variants. This allowed the doctors to begin treatment with corticosteroids and immunoglobulin. Thanks to this highly accurate diagnosis of the problem and pinpointed treatment, the baby is alive and well today. For 48% of the cases the center works on today, supercomputer-powered genetic diagnosis points the way toward a more effective treatment.
The University of Toronto’s SickKids Centre for Computational Medicine uses a supercomputer operating at 107 trillion calculations per second to predict the minute differences between individual children in order to identify the best treatment for each child under their care.
Researchers at the University of Oslo (Norway) are using a supercomputer to help identify the genes that cause bowel and prostate cancer, two common forms of the disease. There are 4,000 new cases of bowel cancer in Norway every year. Only 6 out of 10 patients survive the first five years. Prostate cancer affects 5,000 Norwegians every year and 9 out of 10 patients survive. The researchers are employing the supercomputer to compare the genetic makeup of healthy cells and cancer cells, paying special attention to complex genes called fusion genes.
The Frédéric Joliot Hospital Department (Orsay, France) is using the powerful supercomputer at the French Alternative Energies and Atomic Energy Commission (CEA) in Bruyères-le-Châtel to improve understanding of how tracers used in PET scans for cancer diagnosis distribute themselves through the body. The goals of this research are to optimize PET scan data analysis and, later on, to personalize the PET scan process for each patient in order to produce better outcomes.
Doctors at Australia’s Victor Chang Cardiac Research Institute are using supercomputer-based gaming technology to identify how individuals’ genetic makeups can affect the severity of their heart rhythm diseases. The researchers built a virtual heart, then applied the recorded heartbeats of patients to the digital heart model in order to spot abnormal electrocardiogram signals. The whole process took 10 days using HPC, instead of the 21 years it would have taken with a contemporary personal computer. In other words, this important work would be impractical without the supercomputer.
HPCwire: To a large degree, mechanistic modeling and simulation – beyond compound structure analysis and docking scoring – hasn’t played a large role in the clinic or basic research. Do you think this will change and what will drive the change?
Conway: Modeling and simulation will continue to play a key role in designing a wide array of medical technology products used in clinical practice, from heart pacemakers to diagnostic imaging tools such as MRI and PET scanners. M&S is also crucial for genome sequencing and precision dosing of pharmaceuticals, both of which are important for precision medicine. I think M&S and advanced analytics will go hand-in-hand in this emerging market.
HPCwire: What haven’t I asked that I should?
Conway: Just that precision medicine will be the next market segment IDC adds to the ones we track in our high performance data analysis, or HPDA, practice. Precision medicine will join fraud and anomaly detection, affinity marketing and business intelligence as new segments that are made up mainly of large commercial firms that have adopted HPC for the first time. We forecast that the whole HPDA server and storage market will exceed $5 billion in 2020. Of that amount, about $3.5 billion will come from existing HPC sites and about $1.6 billion will be added to the HPC market by new commercial buyers. Assuming that precision medicine fulfills its promise over the next decade, it is likely to become the single largest market for HPDA, that is, data-intensive computing using HPC resources.
Steve Conway, is research vice president in IDC’s High Performance Computing group where he plays a major role in directing and implementing HPC research related to the worldwide market for technical servers and supercomputers. He is a 25-year veteran of the HPC and IT industries. Before joining IDC, Conway was vice president of corporate communications and investor relations for Cray, and before that had stints at SGI and CompuServe Corporation.