We have put Watson to work in any number of different ways and in any number of different industries. Healthcare, though, was its first real job. It’s gone to medical school, and even studied health insurance. And now Watson is working with the New York Genome Center to launch a pilot that tackles a new medical challenge – glioblastoma.
The most common kind of brain cancer, glioblastoma annually kills 13,000 people in the US alone. As a cancer of the brain, it’s difficult to take tissue samples, for one, so it can’t be examined like most other kinds of cancers. And it moves quickly. Diagnosis to death is on average only 12 months.
All cancers are a disease of the genome. It’s the genome itself that’s progressively changing from normal to abnormal when someone has cancer. When we can determine which genes start to “go bad,” we can better-determine what specific treatment would work to stop it. Therein lies the challenge: How can we better understand what is happening at a genetic level?
The key to glioblastoma’s genetic code is in the human genome. So while we know our cells’ biochemical pathways, it’s also an overwhelming amount of data – billions of DNA base sequences, plus millions of studies, medical documents and clinical records.
Different kinds of brain cancers manifest in different ways and progression rates, so finding these details about glioblastoma is a molecule-sized needle in the genome haystack.
That’s why my team – with decades of research experience in biology as a data science – and NYGC, with the expertise and resources of a dozen top hospitals and medical schools, are collaborating on a project with Watson in genomics. Our goals with this prototype and ensuing studies are to assist physicians with discovering personalized treatment for patients with glioblastoma.
Watson can read millions of pages of medical literature in seconds. By applying its natural language processing and analytics to the genome, it could find connections between what’s buried in journals about the interaction of certain genes, and where those genes are in the genome. And so, in the same way Watson evaluates and hypothesizes on other medical diagnosis based on electronic health records and a doctor’s evaluation (see a demo), it could evaluate and hypothesize about mutations in a cancer cell’s genome that caused the disease, not based on a wide demographic swath of those with similar characteristics, but for an individual based on their personal genome.
Connecting medical literature to the genome
Today, we know and have detailed medical literature on the biochemical pathways our genes take. But we don’t know where in the genome these cancerous perturbations happen in that molecular network of interactions. So, we’re loading Watson with genome data from NYGC, along with medical literature to map out where these deviations happen. Watson will be able to see that, in the context of given cancer mutations in the genome, which pathways matter. And in the context of those interactions, suggest evidence of potential treatments.
IBM Watson and New York Genome Center. Video: IBM SocialMedia
This journey takes clinicians from trials, to validating what genomic knowledge improves treatment, to routine analysis that helps patients. Ultimately, we want to see our partners at NYGC and physicians upload genomic data into the Watson Genome on the cloud, where the system could quickly synthesize a personalized report of available evidence of treatment options.