When IBM ($IBM) was preparing its artificial-intelligence program Watson to compete on the game show "Jeopardy!," it made the system read Wikipedia. The training worked and Watson beat its human competitors. Now IBM and the New York Genome Center (NYGC) are testing whether making Watson read PubMed can help defeat cancer.
NYGC plans to use Watson to deliver more targeted, effective treatments to patients by speeding up analysis of tumor genome samples. The first phase of the project is limited to around 25 patients with glioblastoma, a form of brain cancer. Watson will compare genome data from these patients to medical literature to identify what is causing the cancer and what treatments are likely to be effective. Currently doctors do this work manually, but it is expensive and time-consuming.
"We're hoping Watson's learning model can find associations faster than we can and they'll be able to tune the sets of drugs to at least prioritize and give doctors the ability to drill down so that they can make better determinations of what to try," NYGC's deputy scientific director for informatics, Toby Bloom, told Forbes. The team behind the project described it as part clinical, part research, with the findings uncovered by Watson driving changes in the treatments physicians prescribe.
To prepare Watson for the work, IBM has focused the system's text-analysis capabilities on medical information sources, such as the research abstracts on the National Institutes of Health's PubMed database. Watson was also fed NIH's machine-readable lists of signaling networks and protein interactions to add to its understanding of what goes on inside cells. Over time IBM may add other sources and the full text of journal papers, Ars Technica reports, and Watson will learn what is reliable.
How far IBM develops the system depends, in part, on the success of the initial project. IBM and NYGC have limited themselves to 25 glioblastoma patients for now but may expand the program in the future. The transition of Watson to a cloud-based system--and the nonexclusive nature of the NYGC deal--means IBM could replicate the model with other partners, too.