NEW YORK (GenomeWeb) – The Broad Institute of MIT and Harvard and IBM launched a five-year, $50 million research initiative — funded by IBM — last week that will couple sequencing and genome-editing techniques with IBM Watson's cognitive computing capabilities to study drug resistance in thousands of tumors. They plan to make anonymized data from their studies available to the scientific community for research use.
The American Cancer Society estimates that drug resistance plays a role in nearly 600,000 annual cancer deaths in the United States. "Despite being a big clinical problem, we know remarkably little about what causes tumors to either be resistant to drugs from the beginning or develop resistance over time," Todd Golub, chief scientific officer and founder of the cancer program at the Broad Institute, told GenomeWeb this week. "The vast majority of genomic approaches as part of projects like the [The Cancer Genome Atlas] have focused on newly diagnosed cancers and we know much less about what cancers look like when they are in the drug resistant state."
The new study will be one of the largest to date on the subject, according to the partners. While a number of small studies have been done to try to understand drug resistance in specific cancers, these have focused mostly on hundreds rather than thousands of patients, Kathy McGroddy Goetz, IBM Watson Health's vice president of partnerships and solutions, told GenomeWeb. "[Drug resistance is] a huge scientific problem that is really important to address, and the best way to address it is by getting the appropriate dataset that has enough, hopefully, information in it that Watson will be able to connect all these disparate data points and extract insights and really provide us guidance," she said.
Specifically, the researchers expect to use Watson to analyze data from 10,000 samples from patients recruited from major hospitals in the Boston area that the Broad has collaborations with, patients from several major cancer centers across the country, and patients recruited directly. The partners will try to identify patterns in the data that help researchers and clinicians predict drug sensitivity and resistance. For example, they are planning to look at data collected from a single patient who has experienced drug resistance at multiple time points, McGroddy Goetz said.
Details of how the research plans will proceed over the next few years are still being discussed internally but the researchers plan to do whole-exome sequencing on both tumor tissue and blood samples. "It won't be exactly 2,000 genomes per year for five years because we'll need to ramp up some of the capabilities over that period" but "our goal is to analyze 10,000 cancer genomes over the five years," Golub told GenomeWeb. Also, while the focus is on whole-exome sequencing, if whole-genome sequencing costs drop further in the coming years, the partners may switch to that in the future.
Given the size of the study, there will be a number of logistical challenges, for example around acquiring patient samples. This will be enabled by leveraging new technologies, such as cancer testing from blood, that will make it possible to engage patients even if they don't happen to be local to Boston or be treated at major research cancer centers, Golub said. "There will also be the challenge of not just generating the data but then understanding what it all means, and this is where IBM's computational expertise, combined with other computation approaches, we're optimistic, will make a dent in the problem."
The partners intend to focus their studies on particular tumor types but that list won't be exhaustive. "We do think there will be benefit in having some focus on particular cancer types but we also know that often these types of cancer mechanisms cut across different tumor types and you might miss those if you focus too much," he explained. So "we'll pick a blended approach that will look with some focus but also across several cancer types."
Furthermore, the partners plan to analyze different types of resistance. These include the type of resistance that is seen when tumors fail to respond to initial treatment, as well as acquired resistance, which is when tumors initially shrink but then grow back. "The exact blend of those two types of approaches, I think, we'll learn as we go and [we'll] adapt the study design based on which of those proves to be more informative," Golub said. The Broad is also exploring mechanisms for making deidentified data from patients available in its entirety to the community, although it will most likely be provided through existing repositories, he said.
According to McGroddy Goetz, the current project with Broad represents the next challenge for Watson. Following its performance on Jeopardy, Watson quickly moved into cancer projects with Memorial Sloane Kettering Cancer Center, MD Anderson, and Mayo Clinic, which aimed to train Watson to work with medical doctors and oncologists and help them make recommendations based on current guidelines and identify clinical trials for patients.
These partnerships resulted in the versions of Watson — Watson for Oncology and Watson for Genomics — which are now available to the market. Recently, IBM and Quest Diagnostics launched IBM Watson Genomics from Quest Diagnostics, a service that will be available to oncologists nationwide and that focuses on cancer diagnostics and providing options for evidence-based methods of treating cancer. Broad and Memorial Sloan Kettering Cancer Center are named as partners in that collaboration.
Over the summer, IBM and Hangzhou CognitiveCare announced that 21 hospitals across China plan to adopt Watson for Oncology in order to help their clinicians personalize cancer treatments for their patients. Also, IBM announced a two-year partnership with the US Department of Veterans Affairs to help the agency accelerate its efforts to provide personalized medical care to veterans with cancer.
In March, IBM and the New York Genome Center extended an existing partnership focused on generating and analyzing genomic and clinical data from glioblastoma patients to focus on cancer in general. That effort seeks to analyze sequence data from 200 cancer patients and compare how different types of sequencing approaches help inform and improve cancer care. The partners are using Watson to analyze variant, gene expression, and other data to identify likely driver alterations for the 200 cancer patients as well as to locate actionable and potentially actionable targets for approved, investigational, and off-label drugs.
Now that Watson can use genomic information to tackle treatment recommendations and link patients to clinical trials, "you start thinking, 'What's the next frontier?'" McGroddy Goetz said, and one of the areas that's been getting a lot of attention is drug resistance in cancer patients. The Broad was a "natural" partner for IBM for this project, she said, partly because of its proximity to IBM's Boston offices but in large part because of its expertise in sequencing and because of the software tools it has developed for processing raw sequence data. "We'll see as we go and learn, but it wouldn't surprise me if we leveraged a combination of technologies [for the study] and maybe [even] come up with some new things," she said.
Also, making deidentified data publicly available both benefits the community and improves Watson, she added. "If people around the world are able to leverage this dataset and ... drive new science, the way that will play out is, new scientific articles will be published," she explained. "We can then take those new scientific articles and feed them back into Watson, which makes it smarter and more scalable and [able to] provide additional value to all those doctors that are trying to treat their patients."