Using Watson to Analyze Genomic Data to Personalize Treatment for Patients With Cancer

A Conversation With Robert B. Darnell, MD, PhD

Get Permission

Three years ago, IBM’s Watson supercomputer was best known for defeating two former champions on the TV game show Jeopardy! Today, it is grabbing headlines for becoming an important assistant in cancer care.

Able to read and understand millions of pages of text within seconds, Watson caught the attention of a handful of cancer institutions interested in testing how well the computer could comprehend complex medical information from a patient’s medical record. The medical information included blood test results; pathology and imaging reports detailing tumor type, size, and location; and the presence of genetic mutations. It then offered physicians a list of treatment options, including clinical trials.

Among those institutions is the New York Genome Center, a consortium of 18 medical research centers in New York and New Jersey. The center is now evaluating Watson for Genomics, a cloud-based technology that uses computational biology approaches to analyze patient genomic data, identify actionable mutations, and find the best therapeutic options, in studies of patients with glioblastoma and other cancer types.

Robert B. Darnell, MD, PhD

Robert B. Darnell, MD, PhD

The research is headed by Robert B. Darnell, MD, PhD, Founding Director of the New York Genome Center, Heilbrunn Professor and Senior Physician of Cancer Biology at The Rockefeller University, and Investigator at the Howard Hughes Medical Institute. Dr. Darnell and his colleagues are using whole-genome and RNA sequencing in a clinical research study of 30 patients with glioblastoma to identify genetic mutations in the patients’ tumor and compare the results obtained by expert curation of data and Watson to find the most effective matches with potentially actionable drugs. Dr. Darnell is in the process of finalizing the results from this study and expects to publish them soon.

Dr. Darnell has also launched a project called the Cancer Alliance, a joint study with IBM that has a similar design and includes the goal of creating a national open cancer data repository of genetic data from 200 patients with cancer who have not responded to prior treatments.

The ASCO Post talked with Dr. Darnell about his research and how Watson is accelerating precision treatments for patients with cancer.

Speedy Analysis of Tumor Mutations

How are you using Watson to analyze patients’ genetic information?

We first started using Watson in 2013 in our study of 30 patients with glioblastoma, because it was taking so many people-hours to evaluate the results of very deep sequencing on their tumors, which can involve the discovery of thousands of mutations of different sorts, such as single-nucleotide changes, insertions and deletions in DNA, duplications of whole chromosomes, fusions, and DNA rearrangements. In the initial cases we studied, our team of 12 researchers was putting in at least 160 or more hours per individual patient’s tumor sequencing to filter down their mutations to ones we thought were most likely to be clinically druggable targets.

We became attracted to Watson because of its natural language–processing capability, which can comb medical literature databases and read millions of papers in seconds. It also can analyze patients’ tumor mutations in minutes, a process that typically takes weeks, and then produce a report that may inform treatments.

In our studies using Watson, we are trying to first see how much information we can get using this type of state-of-the-art technology. The standard approach to genomic sequencing currently is to do targeted gene-panel sequencing of the tumor to see what results can be obtained, for example from a panel examining mutations in 400 genes that are thought to be important in a specific cancer. Such panels are technically different from whole-genome sequencing systems we are using, which have the capability to detect with greater sensitivity and specificity the in-depth mutation variants in tumors, particularly when paired with RNA sequencing of the tumor. Watson can quickly analyze these sequence results, and we are testing its speed and accuracy to present drug options for individual patients, which would normally take 2 to 3 weeks to produce.

An Ambitious Plan

You are also using Watson to create the Cancer Alliance, a national tumor registry matching genetic characteristics with available treatment options for patients with a variety of cancers. Is this project part of the National Institutes of Health Precision Medicine Initiative?

What I see in the not-too-distant future is a best-in-class approach to the treatment of cancer, which will require a human and machine interface, because neither alone will be sufficient to give the best possible answers for the treatment of complex cancers.
— Robert B. Darnell, MD, PhD

Tweet this quote

It certainly is 100% in sync with the national Cancer Moonshot effort and the ideas proposed in the Precision Medicine Initiative. We started the glioblastoma project 3 years ago, before the Cancer Moonshot was launched. Like the patients in the glioblastoma study, the patients enrolled in the Cancer Alliance study—about 175 so far—have no conventional treatment options available to them. Their cancers include different types of tumors, including brain, thyroid, lung, skin, and pancreatic cancers. These patients have the greatest need and could benefit the most from genomic analysis.

Our plan is to analyze the data as we are doing for the glioblastoma project and also to aggregate the data from the genomic sequencing of their tumors and put the information in an open cancer data repository. Ultimately, we hope these strategies will help to improve physicians’ ability to determine potential treatment for other patients with cancer mutations.

This is an ambitious plan, and it is the right direction to go in as the data sets get larger and the cognitive computing interface gets better and better. The goal here is to figure out how to scale this technology in a cost-effective but maximal information-leveraged manner. Watson is giving us a tremendous opportunity to get a better understanding of cancer and has the potential to be clinically transformative in the treatment of cancer.

Encouraging Performance Thus Far

You have not published your findings yet from the glioblastoma study. Is there anything you can tell us about your analysis so far?

We are just nearing completion of our analysis and are finding that over the course of the study, Watson seems to be improving in its ability to accurately develop personalized treatment options for each of the 30 patients in the study. The strategy evaluates the genetic mutations present in their tumors and information found in established guidelines and the medical literature and compares those results to the types of calls we are making manually. There is still not a one-to-one correlation between the treatment recommendations Watson is making and our recommendations, but we are encouraged by Watson’s performance as a decision-support system.

The Next Frontier

When do you expect to use Watson in the clinical setting?

What I see in the not-too-distant future is a best-in-class approach to the treatment of cancer, which will require a human and machine interface, because neither alone will be sufficient to give the best possible answers for the treatment of complex cancers. Maybe in the long term, Watson will replace humans in this equation, but I’m not seeing the possibility of that yet—we still need people!

Currently, Watson is successfully being used to provide diagnostic and treatment support. Merging that information with deep genomic analytics is the next frontier. ■

Disclosure: Dr. Darnell reported no potential conflicts of interest. The work was funded in part through philanthropic funds of the NY Genome Center and through support from IBM.