Diagnosis Data

Can a computer find a cure for brain cancer?

Subhashree Madhavan, a bioinformaticist with the National Cancer Institute, likes to talk about brain tumors. And sometimes she likes to show them to people. A PowerPoint presentation Madhavan has prepared about tumor research opens with a close-up of a patient's scalp and skull partially removed, revealing a massive tumor embedded in vein-infused brain tissue. Just the sight of it induces wincing. Over the slide, in big yellow letters, she has written, "Debilitating Brain Tumors."

The next slide is equally shocking. Labeled "Poor Survival Rates," it graphs the chances of survival for brain tumor patients in one institute study. The line plummets sharply and rapidly from the time patients enter the study. The median survival time is less than a year.

A viewer might expect another depressing slide to follow. But instead, Madhavan chooses a Web screen shot of a team of determined-looking medical workers. Again in yellow letters, the slide is labeled "Let's Just Do It." The Web page is the log-in screen for Rembrandt, a powerful new brain tumor analysis tool that researchers believe offers perhaps the best chance to find life-sustaining treatments for brain cancer.

Rembrandt, which stands for Repository of Molecular Brain Neoplasia Data, combines clinical research on actual tumors and data on patients with a wide range of genetic information for several types of tumors. Historically, those two data sets haven't been merged, even though the information from one can greatly inform the other. When combined, they offer some of the best insights into how tumors work and grow, which in turn can lead to more targeted, even unique treatment regimes for patients. "The goal is personalized medicine," says Madhavan, who designed the system.

Rembrandt was conceived to be easy to use, as easy as Google in some instances, she says. A user can type in the name of a particular gene found in brain tumor patients, perhaps one that directs cancer cells to grow, and with a single mouse click, a graph appears, showing survival rates of those patients. With a few more clicks, more charts appear, showing the prevalence of that gene in patients whose histories reside in Rembrandt. A user can narrow down the profiles, filtering for age or gender, for instance. Rembrandt reveals not only how long patients with particular tumors are living, but how well they're responding to specific treatments. The information is more detailed and potentially instructive than anything previously available in cancer research. "It's the entire big picture of trying to make an impact on brain cancer," says Peter Covitz, the core infrastructure director at the cancer institute's Center for Bioinformatics in Rockville, Md.

Rembrandt is something of a coup from a bureaucratic perspective. One reason the different stores of information it houses haven't been merged before is that they weren't shared between doctors and researchers. Even though both sides are keenly interested in finding a cure for cancer, they've performed their work largely in silos. Madhavan and Covitz say that just creating a shared repository has helped foster support for "team science," which they and others believe is required to advance the research field toward a cure.

Rembrandt was built using open-source methods, which are available to anyone, and uses extensible markup language, XML, an easy way to build common information formats. It's also free. In the private sector, where Madhavan and Covitz worked before coming to the National Cancer Institute, intellectual property concerns and profit interests would likely keep the design and use of such a system tightly restricted, Madhavan says. In October, she won a Service to America Medal for Science and Environment from this magazine and the Partnership for Public Service.

New features are being added to Rembrandt, and it needn't be used solely for brain cancer research, Madhavan says. Indeed, her team built a similar application for a breast cancer study; it took only a month and a half to tweak the basic design. The system could be pointed at just about any disease where clinical and molecular data is available. As Madhavan says: "That's the beauty of this informatics portal."