Dr. Ravindra Kolhe turns to his computer screen and uploads raw genetic data sequenced from a very rare cancer directly to an artificial intelligence machine that once competed against and defeated human contestants on quiz show Jeopardy.
After setting up a few parameters – there isn’t even a category that covers the type of cancer he is sending, called desmoplastic small round cell tumor – he opens up some previous reports, but is quickly interrupted.
“OK it’s done,” Kolhe said. Less than a minute after he sent off the raw data, he has back a report from IBM Watson for Genomics that not only identifies the three known cancer-causing mutations in the sample, but 10 clinical trials already underway for drugs targeting those various mutations.
The IBM Watson computer system gained international acclaim in 2011 on Jeopardy when it defeated two former champions head-to-head. But the artificial intelligence system has gone on to provide applications in a number of different industries and fields, including health care. In this case, Kolhe is collaborating with IBM Watson for Genomics in hopes of creating a rapid and comprehensive genetic assessment of patient tumors that could be applied in a timely fashion in the clinic.
Genetic sequencing of tumors is already commonplace at the Georgia Cancer Center, he said. Kolhe has created a 170-gene panel that is run against tumor samples to help identify those mutations already known to cause cancer.
“These are very well established oncogenes and it encompasses pretty much all tumor types,” he said. The problem isn’t getting data, Kolhe said. It is what you do with it.
“One of the major issues is data analysis,” he said. “At least in the clinical space, it is impossible to get people to manually go through that data and do the analysis.”
Kolhe has already partnered with smaller companies to do some analysis, but it is usually limited to a handful of mutations. Enter Watson for Genomics.
Kolhe sent it the raw data sequence for another rare tumor called alveolar soft part sarcoma and, in 20 minutes, got back a comprehensive 31-page report that noted seven “actionable alterations” or known cancer-causing mutations, for which one drug was approved for that particular cancer, five others for other cancers, but targeting those mutations, and 24 clinical trials related to those mutations in various stages of development. It also noted nearly 100 mutations of “unknown significance,” listed in case their relevance is known in the future.
“This is phenomenal,” Kolhe said. “It does the analysis. In the clinical space, I cannot have the time and liberty to go through each and every gene. Getting help from artificial intelligence definitely helps to cut down on time.”
Even if time were not a factor, “I don’t think I can do that manually,” he said. Part of the advantage of using Watson is the number of databases the platform can access, Kolhe said.
“It is amazing the kinds of resources IBM has spread across the world,” he said. “They have access to pretty much everything published, non-published, clinical trials. They have collaborations with pharmaceutical (companies) to have Watson access that information.”
It is not just IBM getting into cancer genome analysis, “but there’s huge investment coming up” in the field, with companies like Google who are said to be looking into it, Kolhe said. There is reason to believe it could have widespread applications once it moves beyond the research and experimental stage.
The American Association for Cancer Research announced last week an analysis of its 19,000-patient sample Project Genomics Evidence Neoplasia Information Exchange (GENIE) found more than 30 percent of those cancers had a known mutation targeted by either a current drug or one in clinical trials.
For his part, Kolhe said he is concentrating on rare cancers and is tecollaborating with Mayo Clinic, which has the largest database of rare cancers. Not only do they demonstrate what the Watson system can truly do, but it is also less likely others are devoting their time and energy to them, he said.
The alveolar sarcoma, for instance, is a rare and often misdiagnosed cancer that spreads in nearly 80 percent of cases and is often resistant to standard chemotherapy, according to The Liddy Shriver Sarcoma Initiative. The desmoplastic sarcoma may be even more rare, with more than 200 cases documented in the medical literature so far.
One of the things Kolhe is hoping to do is run a number of its samples through Watson and “see if we can catch some common denominators and see if that is something very specific for that group,” he said. “That’s the research side of that.”
Even with all of the information Watson provides, including how relevant a clinical trial may be to that particular cancer, each report comes with the disclaimer that Watson “is a research tool. You remain responsible for the conduct of your patient care and for evaluating the clinical relevance of the information provided by the tool.”
And in fact, it is up to the clinician to look at all the options and decide what might be the best options based not just on mutations and drugs, but also what works for the patient in terms of things like geography, Kolhe said. He is still in the process of validating the platform, but hopes to use it for the first patient sometime this summer.
That people will probably already be familiar with Watson, at least part of what it can do, should help, Kolhe said.
“Something like Watson is definitely helpful to bring in patients and convince patients themselves that this is a great tool,” he said.
Reach Tom Corwin at (706) 823-3213 or email@example.com