Framingham Startup Uses Artificial Intelligence To Find Cancer Drugs

Download Audio
A researcher at Berg Health's laboratory in Framingham. (Courtesy of Berg Health)
A researcher at Berg Health's laboratory in Framingham. (Courtesy of Berg Health)

It’s called BPM 31510. It’s one of many cancer drugs being developed in Massachusetts, but it’s different.

BPM 31510 may be the first drug candidate that did not come from a scientist running experiments based on a hypothesis. Instead, it came from artificial intelligence.

The story of how a Framingham company is flipping the scientific method to come up with new drugs like BPM 31510 started about 10 years ago, when Niven Narain was a frustrated cancer researcher at University of Miami's medical school.

At the time, Narain observed physicians who tended to see patients with cancer as disease cases rather than individuals with a biological story to tell.

"Seeing the way the doctors were specifically scripted of treating patients," Narain remembered. "With: ‘You have this disease. This is the first line of therapy. That’s the second line of therapy. If that fails, here’s your options. If that fails, well...’ "

Well, the next option is to try an experimental drug, if there is one. But it takes more than 10 years and more than $1 billion to bring a drug to market. Narain thought that was too long and too expensive. And he convinced billionaire Carl Berg — someone who'd vowed never to invest in the pharmaceutical industry because the runway was so long and risky — to help him find a faster, cheaper way.

Inside the Berg Health facility here in Framingham, you see all the things you’d expect to see in a pharmaceutical lab, including centrifuges and scientists in white coats. There's also a room full of high-throughput mass spectrometers, sophisticated chemical analysis instruments that together cost more than the average American earns over a lifetime.

"These machines are running 24/7, every day of the year," Narain said.

They're processing tissue, urine and blood samples, both from thousands of cancer patients and from thousands of healthy patients, many samples that were taken over long periods of time. Narain says the machines are churning out huge amounts of valuable raw data.

Freezers are the Fort Knox of Berg Health's facility in Framingham. They store valuable samples of thousands of patients over time, that help the company read the story of an unfolding cancer to try to find ways to stop it. (Curt Nickisch/WBUR)
Freezers store valuable samples from thousands of patients. These samples help Berg Health read the story of an unfolding cancer to try to find ways to stop it. (Curt Nickisch/WBUR)

"The last thing you want to do now," Narain said, "is have a hundred biochemists, you know, going through this data and saying, ‘Oh, I kind of like this one over here, or I think this…’ That would make us no different."

What makes Berg different is that it built an artificial intelligence system. The complex computational platform processes trillions of data points to identify molecules that may be effective drugs.

"So we’re not dismissing the scientific process," Narain said. "We’re just kind of putting one step ahead of it. Instead of the human brain deciphering or proposing those hypotheses, we’re allowing the data from the patient biology to generate those hypotheses."

Working at the rare Boston biotech company that integrates mathematics and computers so prominently has been a blast for Slava Akmaev, Berg Health’s chief analytics officer. But Akmaev says the company's biologists and other traditional lab researchers don’t mind that computational scientists like him are driving the experimental agenda.

"No, I don’t think it’s about being elevated or not," Akmaev said. "I think it’s more about trying something new."

And at least so far, trying something new seems to be working. Berg’s computational tools have identified drug candidates for different cancers and for Alzheimer’s and Parkinson’s diseases. And they've done so in a relatively short amount of time.

Computational tools help Berg Health pinpoint meaningful links between different molecules in samples from cancer patients. (Courtesy)
Computational tools help Berg Health pinpoint links between different molecules in samples from cancer patients. (Courtesy of Berg Health)

"To have that many drug and pipelines going successfully is amazing to me," said Atul Butte, the incoming head of the new Institute of Computational Health Sciences at the University of California, San Francisco. Butte says Berg is just one of a number of companies trying their hand at data-driven discovery of new therapies. He says it’s very complex work with plenty of potential pitfalls, but that the 200-person Framingham company is off to an impressive start.

"You have all this genomic data plus all these other molecular measurements," Butte said. "All this computational work to figure out what’s causing what. And if that works, they’re going to be the first to come up with the first data-driven drugs."

But the first data-driven drug on the market is still a long ways away. Berg Health's most advanced drug candidate, BPM 31510, is in early human trials. Even so, Narain feels good about how far Berg has come.

"The way that we get to these potential candidates may be a bit disruptive and a bit different," the president and co-founder said. "But they’re also leading us into areas of biology that are completely unknown. And that is actually tremendously exciting."

This segment aired on March 17, 2015.


Headshot of Curt Nickisch

Curt Nickisch Business & Technology Reporter
Curt Nickisch was formerly WBUR's business and technology reporter.



More from WBUR

Listen Live