The HPC Impact showcase provided a brilliant talk by Rick Stevens from Argonne National Laboratories. These are my notes – as usual, stream of consciousness, all mistakes are almost certainly mine.
Rick gave an excellent intro to what cancer is and why it’s a problem. I am leaving out the technical details of the DOE/NCI collaboration he spoke about, and trying to focus on what this kind of research means for you.
What is Cancer?
Large number of complex diseases, each behave differently depending on the cell type it originates from, and there are many factors that influence the disease – age of onset, invasiveness, responsiveness to particular treatments.
It is basically abnormal cell growth where the body can no longer regulate growth. It spreads when pieces of the tumour break off and spread to other parts of the body – metastasis.
Cells that start to grow abnormally normally are destroyed by self-inflicted cell death or the immune system, but sometimes cells bypass these death signals and continue to grow.
Cancers acquire the ability to ignore cell growth regulators, avoid the immune system, and grow limitlessly, that make them catastrophic for our systems.
Why is cancer hard? Because bodies accumulate mutations over time and it usually takes 2 or 3 mutations before tumour supression genes get deactivated or oncogenes get activated. They disable DNA repair mechanisms so that mutations occur at a faster rate. As you get older you get accumulated mutations and very complex behaviours. Chemotherapy can work but often some subset of cells will eventually become resistant due to high mutation rates.
Tumours develop drug resistance due to massive heterogeneity. No one drug will target all of the varied cells. So different cells in the same tumour will respond differently to the chemotherapy or other treatment.
Precision oncology is about personalized cancer therapy. The population of cancer patients is not homogeneous and neither are the cancers, so take personal measurements of both person and tumour and apply models that predict how well specific treatments will work for particular patient profiles. Using molecular data to assign treatment based on likelihood of success.
One of the motivators for the DOE NCI joint project in this space is that new cancer drugs on average cost $100,000 per year per treatment, and the average survival improvement is only 6 months. That’s pretty mind boggling.
So the impact of High Performance Computing precision oncology research is to create systems that will take blood tests, biopsy results, and maybe even images from CAT scans and the like, and produce an accurate picture of which drugs are most likely to work for you.
At the moment, despite the ways described above in which individual cancers, and even individual cells within cancers, can vary, our treatments are largely uniform. If you have cancer A you will get drug X. And if that doesn’t work they will move on to drug Y. etc. With these new systems – which are really close to practical implementation in the clinic – your doctor can feed your test results into the software, and it will spit out the probability that each of the available treatments will actually work.
HPC and precision oncology could save your life. What more proof do you need that #HPCMatters?