One of the most promising advances to come out of the intersection of HPC and life sciences is personalized medicine. A new video from Intel highlights the current state of personalized medicine and identifies some of the main technical barriers that still constrain its adoption.
As put forth by Christopher Gough lead solutions architect in the Intel Health & Life Sciences Group in a recent blog post, “the goal of personalized medicine is to shift from a population-based treatment approach (i.e. all people with the same type of cancer are treated in the same way) to an approach where the care pathway with the best possible prognosis is selected based on attributes specific to a patient, including their genomic profile.
“After a patient’s genome is sequenced, it is reconstructed from the read information, compared against a reference genome, and the variants are mapped; this determines what’s different about the patient as an individual or how their tumor genome differs from their normal DNA. This process is often called downstream analytics (because it is downstream from the sequencing process).”
Although the cost of sequencing has fallen dramatically over the last decade to enable the long-anticipated $1,000 mark, the cost of interpreting genomic data will become the limiting factor. This requires sophisticated analytics software to identify relevant genotypes and analytics algorithms to guide physicians in choosing the right therapies based on the patient’s specific profile. As Gough observes, the cost of delivering personalized medicine in a clinical setting “to the masses” is still quite high.
Gough identifies the following technical barriers:
- Software Optimization/Performance: While the industry is doing genomics analytics on x86 architecture, much of the software has not been optimized to take advantage of parallelization and instruction enhancements inherent with this platform
- Storing Large Data Repositories: As you might imagine, genomic data is large, and with each new generation of sequencers, the amount of data captured increases significantly. Intel is working with the industry to apply the Lustre (highly redundant/highly scalable) file system in this domain
- Moving Vast Repositories of Data: Although (relatively) new technologies like Hadoop help the situation by “moving compute to the data”, sometimes you can’t get around the need to move a large amount of data from point A to point B. As it turns out, FTP isn’t the most optimal way to move data when you are talking Terabytes
When it comes to saving lives, every second counts. In the case of one pediatric oncology center, adopting Intel technology reduced their genomics analysis time from seven days to four hours. This means that patients can be sequenced multiple times during the course of treatment enabling clinicians to detect signs of drug resistance and offer tailored treatment.
Another team of researchers working to develop new drug treatments, provisioned a cloud cluster of 169,600 Intel Xeon cores, screening ten million compounds in just eleven hours, resulting in three promising target compounds. This represents 39 years of science and would have cost $44 million if a traditional infrastructure had to be purchased and operated.