How biotech companies are using AI to design drugs

Editor's Note: This story was featured in WBUR's weekly health newsletter, CommonHealth. If you like what you read and want it in your inbox, sign up here.

One way artificial intelligence promises to revolutionize health care is in the realm of drug development. Creating new treatments for diseases is an expensive, time-intensive process that typically involves lots of failure, sometimes over a decade or longer. Most drugs that appear promising in the lab never make it to pharmacy shelves.

In some ways, it’s a testament to the complexity of the human body. Once new treatments get inside us, more often than not, they fail to function as researchers expected.

Computers — even super smart ones — aren’t likely to change that, at least not yet. But they may be able to help researchers reimagine the drug discovery process and target ideas that are more likely to succeed.

That’s the hope, anyway. And many biotech and pharmaceutical companies are already diving in. They — and their investors — are betting the newest generation of AI can help unleash novel treatments for everything from COVID to cancer and chronic diseases.

Generate:Biomedicines, a Somerville-based company, has trained its AI on the amino acid sequences that make up proteins — essentially the code that drives much of biology. Now, the company’s scientists are using AI to design new proteins that don’t exist in nature. The idea is to expand exponentially the universe of proteins with disease-treating potential.

Generate was founded roughly five years ago with backing from Flagship Pioneering, the same company that helped spawn Moderna. Since then, it has generated about 5 million proteins “that nature hasn’t discovered,” said CEO Mike Nally.

“We've built all those, we've tested all those and we've learned from all of those to refine our computational models,” Nally said. “And what's really exciting is that the answers are just getting better and better.”

The company hopes to find treatments for illnesses like autoimmune conditions, cancer and infectious disease. Two of its drugs are now entering the first phase of clinical trials: one for asthma and a COVID-19 monoclonal antibody treatment.

“These sort of technologies, I think, will allow us to provide a better line of defense against nature's greatest threats.”

Mike Nally

The COVID treatment targets a spot on the coronavirus’s spike protein that seems to retain its shape as the virus mutates. If it works, the drug could provide an added layer of protection for people whose immune systems are compromised and vulnerable to illness even after vaccination. Nally said the study is fully enrolled, and all 50 people have received the infusions without major safety concerns.

“Our COVID program was 17 months from concept to clinic,” Nally said, compared with an average of five to six years for comparable treatments. He sees this as a sign of the enormous potential for AI.

“It’s going to be faster in the future,” Nally said. “These sort of technologies, I think, will allow us to provide a better line of defense against nature's greatest threats.”

In the area of infectious diseases, the company also hopes to tackle flu.

Cambridge-based Montai, another company affiliated with Flagship Pioneering, is focused on molecules that already exist in nature, specifically in foods, traditional medicines and other substances that humans consume regularly.

The company is combing these molecules for clues that could lead to daily pills for chronic diseases, like inflammation.

People have looked at this area before, searching places like the Amazon for plants with medicinal benefits, explained Montai CEO Margo Georgiadis. But it was difficult to sort the data systematically at a large scale, and try to isolate individual molecules and what they do in the body.

“These are things that we know have a [human] biological benefit,” Georgiadis said. “I call them hidden in plain sight — keys that can connect to our biology, but we haven't been able to look at and tour this landscape before.”

Montai spent several years teaching its AI about the chemistry of these substances and about human biology. Then, it used the AI to sort through the molecules and predict which ones would be most likely to target specific pathways in the body. Georgiadis said it has identified more than 125,000 substances that show potential to be refined into treatments for chronic illnesses.

“We believe that they’re some of the richest starting points for creating safer therapies for chronic disease,” she said.

Oral drugs like the ones Montai is developing are generally less expensive and easier for patients to manage than injections or infusions. But the success rate in the clinic for traditionally developed drugs of this type is low — under 10% according to Georgiadis. Montai’s scientists believe AI — with its capacity to handle enormous data sets and look for patterns and connections — can help them do better, and reduce the cost of developing these drugs.

The real proof will come after clinical trials, which will show whether drugs designed with help from AI tools are as safe and effective as scientists hope. Montai is about a year and a half away from putting some of its first drugs into clinical trials, Georgiadis estimated.

For now, the promise of AI in drug development remains unproven. These systems and the humans training them have yet to put a single treatment into pharmacies, said Jen Nwankwo, CEO of 1910 Genetics, a biotech company headquartered in Boston’s Seaport District.

“In the first iteration, it might not be cheaper than a traditional process, or it might be cheaper, but not faster. ... And that's just the way technology evolves.”

Jen Nwankwo

“We just need to put drugs on the shelves,” Nwankwo said. Until then, it's hard to know whether AI is more efficient and less expensive than traditional drug development.

“In the first iteration, it might not be cheaper than a traditional process, or it might be cheaper, but not faster," she said. "It might be faster, but not cheaper. And that's just the way technology evolves.”

1910 Genetics started in 2018 and has created an AI model aimed at developing both oral drugs and infusions. One of its first targets is a pain relief alternative to opioids.

The company, which has partnered with Microsoft, also plans to make its AI system available to other drug developers.

Nwankwo emphasized that AI will not discover drugs on its own. The process still requires plenty of work by humans. And there are significant challenges, including the fact that AI is only as good as the data people provide. There are still many gaps in our knowledge.

The first AI-assisted treatments to hit pharmacies may be a few years away, said Generate:Biomedicines’ Nally. But 2024 is shaping up to be “an important watershed moment” for the industry, said Nwankwo.

A handful of AI companies expect to report results from the second phase of clinical trials, which measure whether an experimental drug shows enough success treating disease to continue testing.

Over time, Nwankwo said she believes AI will "hit the trifecta of being better, faster, and cheaper," and it will open up new avenues to treat a wide variety of diseases.

“Because if you're a patient,” Nwankwo said, “you want every possible option in your toolbox.”