MONROVIA, Calif. – Terray Therapeutics’ laboratory is a symphony of miniaturized automation. Robots buzz and deliver small tubes of liquids to their stations. Scientists in blue coats, sterile gloves and safety goggles monitor the machines.
But the real action takes place at the nanoscale: combining proteins in solution with chemical molecules in tiny wells in custom-made silicon chips that resemble microscopic muffin tins. Every interaction is recorded, millions upon millions per day, generating 50 terabytes of raw data every day – the equivalent of more than 12,000 movies.
The laboratory, about two-thirds the size of a football field, is a data factory for drug discovery and development using artificial intelligence in Monrovia, California. It’s part of a wave of young companies and startups trying to leverage AI to produce more effective drugs faster.
Sign up for The Morning newsletter from the New York Times
The companies are using the new technology – which learns from vast amounts of data to generate answers – to try to reshape drug discovery. They are shifting the field from painstaking artisanal craft to more automated precision, a shift powered by AI that learns and gets smarter.
“Once you have the right kind of data, the AI can work and become really good,” said Jacob Berlin, co-founder and CEO of Terray.
The majority of early business applications of generative AI, which can produce everything from poetry to computer programs, were intended to help take the hassle out of routine office tasks, customer service and code writing. Still, drug discovery and development is a huge industry that experts say is ripe for an AI makeover.
According to consultancy firm McKinsey & Co, AI is a “once-in-a-century opportunity” for the pharmaceutical sector.
Just as popular chatbots like ChatGPT are trained with text from the Internet, and image generators like DALL-E learn from vast amounts of photos and videos, AI for drug discovery relies on data. And it is highly specialized data: molecular information, protein structures and measurements of biochemical interactions. The AI learns from patterns in the data to suggest potentially useful drug candidates, like attaching chemical keys to the appropriate protein locks.
Because AI for drug development is driven by precise scientific data, toxic ‘hallucinations’ are much less likely than in more broadly trained chatbots. And any potential drug must be extensively tested in laboratories and in clinical trials before being approved for patients.
Companies like Terray are building large high-tech labs to generate the information to help train the AI, allowing for rapid experimentation and the ability to identify patterns and make predictions about what might work.
Generative AI can then digitally design a drug molecule. That design is translated into a physical molecule in a rapid automated laboratory and tested for its interaction with a target protein. The results – positive or negative – are recorded and fed back to the AI software to improve the next design, speeding up the overall process.
Although some AI-developed drugs are in clinical trials, it is still in its infancy.
“Generative AI is transforming the field, but the drug development process is messy and deeply human,” said David Baker, biochemist and director of the Institute for Protein Design at the University of Washington.
Drug development has traditionally been an expensive, time-consuming, volatile endeavor. Studies on the costs of designing a drug and navigating clinical trials to final approval vary widely. But the total cost is estimated to average $1 billion. It takes 10-15 years. And nearly 90% of drug candidates submitted to human clinical trials fail, usually due to lack of efficacy or unforeseen side effects.
The young AI drug developers aim to use their technology to improve those odds, while saving time and money.
Their most consistent source of funding comes from the pharmaceutical giants, which have long served as partners and bankers for smaller research companies. Today’s AI medicines are typically focused on accelerating preclinical development phases, which typically take four to seven years. Some may try to do clinical trials themselves. But at that stage the big pharmaceutical companies usually take over and conduct the expensive human trials, which can take another seven years.
For established pharmaceutical companies, the partner strategy is a relatively cheap way to tap into innovation.
“For them, it’s like taking an Uber to go somewhere, instead of having to buy a car,” said Gerardo Ubaghs Carrión, a former biotech investment banker at Bank of America Securities.
The big pharmaceutical companies pay their research partners to achieve milestones toward drug candidates, which can reach hundreds of millions of dollars over the years. And if a drug is ultimately approved and becomes a commercial success, there is a stream of royalty income.
Companies such as Terray, Recursion Pharmaceuticals, Schrödinger and Isomorphic Labs strive for breakthroughs. But there are roughly two different processes: those that build large laboratories and those that do not.
Isomorphic, the drug development spinout of Google DeepMind, the tech giant’s central AI group, believes that the better the AI, the less data is needed. And it’s betting on its software skills.
In 2021, Google released DeepMind software that accurately predicted the shapes into which strings of amino acids would fold as proteins. These 3D shapes determine how a protein functions. That was a boost to biological understanding and useful in drug discovery, since proteins drive the behavior of all living things.
Last month, Google DeepMind and Isomorphic announced that their latest AI model, AlphaFold 3, can predict how molecules and proteins will interact – a further step in drug design.
“We are focusing on the computational approach,” says Max Jaderberg, Chief AI Officer at Isomorphic. “We think there is enormous potential that can be unlocked.”
Terray, like most drug development startups, is a byproduct of years of scientific research combined with more recent developments in AI.
Berlin, who earned his doctorate in chemistry from Caltech, has made advances in nanotechnology and chemistry throughout his career. Terray grew out of an academic project that began more than a decade ago at the City of Hope cancer center near Los Angeles, where Berlin had a research group.
Terray focuses on developing small molecule drugs, essentially any drug that a person can take in a pill, such as aspirin and statins. Pills are easy to take and cheap to produce.
Terray’s sleek labs are a far cry from the old days in academia, when data was stored in Excel spreadsheets and automation was a distant goal.
“I was the robot,” recalls Kathleen Elison, co-founder and senior scientist at Terray.
But by 2018, when Terray was founded, the technologies needed to build the industrial-style data lab were rapidly advancing. Terray has relied on advances from third-party manufacturers to create the microscale chips that Terray designs. The labs are filled with automated equipment, but almost everything is custom-made, made possible by advances in 3D printing technology.
From the start, the Terray team recognized that AI would be crucial to understanding data storage, but the potential for generative AI in drug development only became apparent later – however, before ChatGPT became a breakthrough in 2022.
Narbe Mardirossian, a senior scientist at Amgen, became Terray’s chief technology officer in 2020 – in part because of the wealth of data generated by the lab. Under Mardirossian, Terray has built its data science and AI teams and created an AI model for translating chemical data to math, and back again. The company has released an open source version.
Terray has collaboration agreements with Bristol Myers Squibb and Calico Life Sciences, a subsidiary of Google’s parent company Alphabet, which focuses on age-related diseases. The terms of those deals are not being disclosed.
To expand, Terray needs more money than its $80 million in venture capital, said Eli Berlin, Jacob Berlin’s younger brother. He left a job in private equity to become co-founder and chief financial and operating officer of the startup, convinced the technology could open the door to a lucrative business, he said.
Terray is developing new drugs for inflammatory diseases, including lupus, psoriasis and rheumatoid arthritis. The company, Jacob Berlin said, expects to have drugs in clinical trials by early 2026.
Terray and his colleagues’ drug manufacturing innovations can speed things up, but only to a limited extent.
“The ultimate test for us, and for the field in general, is whether you can look back in 10 years and say that the clinical success rate has gone up a lot and that we have better medicines for human health,” Berlin said.
c.2024 The New York Times Company