Clarke was inspired to search for cancer stems in 1996 after puzzling over a slide of a testicular cancer patient. The patient was essentially cured, even though he still had a few mature cancer cells left. Suddenly Clarke had an epiphany: "I said, 'Cancer is a stem cell disease.'"

In 2003 he shocked researchers by isolating stem cells inside breast tumors, the first time such cells had been found in a solid tumor. These stem cells formed 5% of the cells inside the tumor, but if just a couple hundred of them were implanted in mice, new tumors took hold. When he implanted tens of thousands of regular breast cancer cells in mice, nothing happened. The finding was so radical, "most people thought I was nuts," he recalls.

But since then cancer stem cells have been found in brain, colon, head and neck, pancreatic and skin tumors. "At least now we know the face of the true enemy," says renowned cancer biologist Robert Weinberg at the Massachusetts Institute of Technology.

The commercial implications were so obvious that Clarke and collaborator Max Wicha sent their breast cancer data to venture investors at the same time they sent it out to be published. The proposal eventually landed on the desk of former Genentech biologist Laurence Lasky, now a partner at U.S. Venture Partners. He was stunned by the data and felt it was too big an opportunity to pass up. "If it is true and I don't do it, I will feel like the biggest idiot of all time,""he recalls thinking. OncoMed has raised $103 million in venture funding to date.

It will be a few years before Clarke knows whether the bet will pay off; OncoMed's first drug hits human trials this year. Some scientists are skeptical of the cancer stem cell theory. But Clarke is confident. "We are just starting to scratch the surface," he says.