New York Times,
March 27, 2014Link
When Robert B. Darnell was a graduate student in the early 1980s, he spent a year sequencing a tiny fragment of DNA. Now Dr. Darnell is an oncologist and the president of the New York Genome Center, where the DNA-sequencing machines can decode his grad-school fragment in less than a ten-thousandth of a second.
As an oncologist, Dr. Darnell is firmly convinced that this technological advance will change how cancer is treated. “It’s inspiring for me, and it’s inspiring for lots of doctors,” he said in an interview.
The idea is simple. Oncologists will get a tumor biopsy and have its genome sequenced. They will identify the mutations in the cancer cells, and they will draw up a list of drugs to treat each patient’s particular mix of mutations.
This isn’t pure science fiction. Oncologists have already created such drug cocktails for a handful of cancer patients. But that doesn’t mean people with cancer should expect personalized treatments any time soon. Unfortunately, the path from a genome to a treatment is blocked by a colossal bottleneck.
“We know that the devil’s in the details, and a lot of the mystery is still there,” said Dr. Darnell.
The problem is that many mutations in a cancer cell may be harmless, and targeting them will be a waste of time. And once oncologists narrow down their list to the mutations that actually drive cancer, they have to understand how they do so. Some mutations cause cells to ignore signals to stop growing, for example, while others promote the growth of blood vessels nearby.
Scientists have already found out a fair amount about those effects. But it’s hard to assemble this existing knowledge. “There isn’t any single magic database,” said Toby Bloom, the deputy scientific director at the New York Genome Center.
Rather than just typing in a mutation into a search box, doctors have to comb the scientific literature. Not only do they have to read many papers, but they have to evaluate the strength of their conclusions.
“You can do it, but it’s not scalable,” said Dr. Bloom.
Some cancer scientists are trying to open the bottleneck with the help of computers. They’re writing software that can read scientific reports and glean their insights.
The New York Genome Center has now joined this movement. It’s enlisting Watson, a computer developed by IBM that first came to fame in 2011 when it defeated human contestants on the game show “Jeopardy.”
Since its television victory, Watson has turned its attention to medicine. Its reading list include the study abstracts stored in Medline, a federal government index. “I believe at this point it’s up to 23 million or so now,” said Ajay K. Royyuru of IBM’s Thomas J. Watson Research Center in Yorktown Heights, N.Y. “We’re taking all of it.”
For a pilot study, Dr. Darnell and his colleagues will use Watson to help them come up with personalized treatments for a type of brain cancer known as glioblastoma. They chose the disease because forms of it are so devastating.
For the worst types, Dr, Darnell said, “It’s as close to a death sentence as you can get,” with patients typically surviving a year after diagnosis. In those cases, surgery, radiation and chemotherapy sometimes help, but they’re hardly a panacea. “It works for about a third of the patients, and it gives them about two extra months,” said Dr. Bloom.
In the new study, the scientists will initially recruit 20 patients. They will sequence genomes from tumor biopsies and feed that data (along with the sequences of healthy cells from the patients) to Watson.
Watson will identify the mutations in the tumor and draw on its medical knowledge to develop a hypothesis for how they cause cancer. It will then put together a list of drugs that could potentially treat the cancer.
“All of this is being done in seconds to minutes,” said Dr. Royyuru.
Those drugs may directly target the mutant proteins in cancer cells, or they may target other proteins they work with. Some of the drugs may have been studied for other types of cancer before, but not for glioblastomas. Some of the drugs may have never been tried out on any cancer before.
A group of experts including neurologists and pharmacologists will look over Watson’s suggestions and make the final decision about whether to give any of the drugs to a patient. Rather than patients being given a single drug, as happens in conventional treatment, Dr. Darnell hopes that in the new study, they will get as many as five drugs at once.
Because glioblastoma is such a deadly disease, Dr. Darnell also hopes he won’t have to wait long to see if Watson helps prolong people’s lives. And Dr. Royyuru will feed the results of the trials back to Watson, so that it can generate new hypotheses about glioblastoma and improve its suggestions.
Other scientists want to see those results before passing judgment on the new project. But they generally agree that powerful computers of some kind will be essential for making sense of cancer genomes.
“This is the course of how things will move forward,” said Kelly A. Frazer of the University of California, San Diego.
Still, Heidi L. Rehm, a molecular geneticist at Brigham and Women’s Hospital, said that computers alone would not deliver personalized cancer treatment. “It’s necessary — but it’s not sufficient,” she said. “It’s only as good as the data going in.”
Watson, for example, is reading only abstracts of papers for now, rather than the entire papers. And the quality of the information in those papers varies tremendously.
Dr. Rehm believes that medical research has to move beyond the tradition of scientific papers. Scientists need to organize their data so that it can go into easily searched databases, such as the Clinical Genome Resource, which she is helping to develop.
No matter how good computers get at understanding medicine, though, cancer researchers don’t think machines will have the last word on how to treat patients. “At some point a doctor has to look at the data and see if makes sense,” said Sameek Roychowdhury of Ohio State University. “That part is never scalable.”
Copyright 2014 The New York Times Company. Reproduced with permission.