From Code to Cure

Armed with enormous amounts of clinical data, teams of computer scientists, statisticians, and physicians are rewriting the rules of medical research.

by David J. Craig Published Spring 2018
  • Comments (0)
  • Email
  • ShareThis
  • Print
  • Download
  • Text Size A A A


It has been nearly forty years since scientists discovered that cancer is caused by flaws in our DNA, and that insight still guides most oncology research today, inspiring scientists to hunt for cancer-causing genes and to search for drugs that help people with particular mutations.

Andrea Califano, the founding director and chair of CUIMC’s Department of Systems Biology, has taken a different approach to studying the disease.

Rather than relying on genetic mutations as signposts in his quest to understand cancer, Califano has plunged headlong into the messy interior dynamics of cancer cells, attempting to determine how the tens of thousands of proteins operating inside cells can conspire to make them divide uncontrollably. It is an approach that has required him to build one of the most complex, data-intensive mathematical models of cellular activity in existence — yet it is revealing that cancer may be a simpler and more treatable disease than we first thought.

“What my team is doing is akin to dismantling a car that’s broken down and then rebuilding it, one piece at a time, in hopes of diagnosing the problem,” says Califano, a former theoretical physicist who worked for several years as a computational biologist at IBM’s Thomas J. Watson Research Center before coming to Columbia in 2003. “We think this may be the only way we’ll ever truly understand how a cancer cell works.”

Califano set out on this path about ten years ago, when cancer researchers were beginning to realize, after years spent hoping that the Human Genome Project would produce a clear road map for fighting cancer, that the disease involves far more genes than anyone had previously imagined. Although a handful of genetic mutations wield a strong influence in causing some types of cancer — thereby giving researchers clues to developing new, personalized treatments — most forms of the disease turn out to involve dozens, or even hundreds, of mutations, each contributing a small portion of a person’s overall risk. To make matters more confusing, the genes at the roots of cancer vary considerably from one person to the next, even among people whose tumors start in the same organ and otherwise look identical.

“So this raised the question: is cancer not one disease but actually thousands of different diseases that we’d have to cure individually?” says Califano. “My hunch, and my hope, was that this wasn’t the case. I still believed there had to be some common cellular mechanisms shared by many cancers that we just hadn’t noticed yet. And I thought to find them, we’d have to look beyond genes — straight into the guts of the cell.”

To many biologists, this seemed like an exercise in futility. No practical methods of studying the inner dynamics of entire cells existed at the time; biologists who studied interactions among proteins therefore restricted their analyses to small groups of molecules extracted from cells. Moreover, many biologists thought that diseased cells would be especially difficult to study, since their interior mechanics were going haywire.

“I never bought that idea,” says Califano. “Maybe it’s my background as a physicist, but I tend to assume that nature is operating as efficiently as possible unless evidence tells us otherwise. I saw no reason to suspect that cells with virtually identical capabilities of spreading rapidly throughout your body aren’t operating in an extremely orderly and consistent manner.”

It turns out that he may be right. In a series of stunning papers published over the past few years, Califano and several members of his lab have identified dozens of proteins that they say act as “master regulators” in cancer cells, seamlessly orchestrating the activities of hundreds of other proteins, which, in turn, force the cells to divide and persist in a malignant state. Califano’s team has accomplished this using a sophisticated investigative strategy, which involves measuring the activity levels of all the proteins in large numbers of healthy and cancerous cells; determining which proteins are capable of binding to one another; mapping out all their potential relationships in gigantic sunburst-shaped charts; and then training a computer algorithm to identify which proteins are most influential in making a cell cancerous. It took one of the largest supercomputers in the world, built under Califano’s oversight at CUIMC in 2008, to perform the calculations.

The therapeutic implications of these discoveries could be profound. Califano says that the cancer-driving proteins that he and his colleagues have identified are active in certain subsets of people with many different types of cancer — an assessment based on their analysis of cells drawn from more than twenty thousand patients from across the United States. The researchers have also conducted experiments on mice to determine which of approximately 120 FDA-approved drugs and 340 experimental compounds are most effective against cancer cells that contain heightened levels of these proteins; based on the results, they’ve developed a computer-based diagnostic system that recommends treatment strategies for cancer patients who test positive for the proteins.

“Often, the recommendations are for drugs that no physicians would have ever even thought to use for a certain kind of cancer,” says Califano. “The system can reveal that someone with brain cancer needs the same medication as someone with lung cancer or someone with leukemia. This is because some of the master regulators we’ve identified crop up in all sorts of cancers that nobody knew had underlying similarities.”

To date, Califano’s diagnostic technology has been used in only a handful of cases, when terminally ill cancer patients in the final stages of the disease sought experimental treatments. But the results have been so promising — with some patients having had their lives extended by six months or longer — that the FDA recently approved a clinical study in which dozens of men and women with pancreatic cancer will have their protein levels assessed by Califano’s team during their initial phase of treatment. Califano and his colleagues will then identify a handful of drugs that might help each patient and then work closely with scientists in the laboratory of CUIMC pancreatic-cancer specialist Kenneth Olive to test their effectiveness in mice that have been injected with the patient’s own cancer cells.

“While we’re performing these individually tailored experiments on mice, the patients will receive traditional care,” Olive says. “And then, based on the response of a person’s mouse avatar, we will select which one, among a dozen additional drugs, should be given to the patient.”

Califano hopes the technology, if it proves successful, will be widely used one day in conjunction with DNA tests — thus marrying the best cancer diagnostics of the genetics era and the emerging age of high-powered protein analysis.

“The best cancer care is going to result from bringing together genetics, proteomics, and other novel approaches like immunotherapy,” Califano says. “We must embrace cancer as a highly complex disease and throw everything we have at it.”


Read the related article Using Data for Good:

  • Email
  • ShareThis
  • Print
  • Recommend (14)
Log in with your UNI to post a comment

The best stories wherever you go on the Columbia Magazine App

Maybe next time