AI is coming. Is the chemistry world ready?

Artificial intelligence, says futurist Jamie Metzl, is soon to go the way of electricity: impossible to unravel from the fabric of everyday life.

The same could be said for artificial intelligence (AI) in drug discovery and development, according to the biotech and pharmaceutical executives and investors who congregated in San Diego for the Biotechnology Innovation Organization (BIO) 2024 International Convention­—Metzl among them. The software is increasingly being put toward the work of identifying targets, designing molecules large and small, and trimming down the time it takes to put together a data package for the US Food and Drug Administration. The FDA itself recently launched its first-ever committee dedicated to digital health and AI.

“AI won’t replace chemists, but chemists that know AI will replace those that don’t,” Stacie Calad-Thomson, business development lead for Nvidia’s health-care and life sciences division, said during a panel discussion on Tuesday.

The value of AI isn’t new to Karen Akinsanya, president of research and development in computational chemistry firm Schrödinger’s therapeutics division. Schrödinger was among the first companies to use software—specifically, a physics-based platform that predicts the binding and structural properties of small molecules—to design drugs that have since proved successful in human studies, including the TYK2 inhibitor TAK-279. Pharma firm Takeda acquired TAK-279 from Nimbus Therapeutics in 2023 for $6 billion. In a midstage study testing its use in psoriatic arthritis, TAK-279 alleviated joint pain and improved function, per data that Takeda presented at the American College of Rheumatology’s annual conference in November.

Akinsanya says the success of molecules like TAK-279 in the real world—not just on a computer—has made it easier to strike pharmaceutical partnerships, including with Bristol Myers Squibb and Eli Lilly and Company. That model is now making its way to smaller biotech firms, too, which Akinsanya says are increasingly relying on predictive software rather than employing “lots and lots of medicinal chemists.”

“We’re seeing lots and lots of biotechs not build large chemistry teams, but just go and buy the platform at pretty interesting levels, actually, that compete with large pharma,” Akinsanya says.

Schrödinger also recently started working on its own pipeline of cancer drug candidates. Akinsanya joined the firm six years ago with the express mission of turning it into a drug company, not just a platform partner. Other early biotech AI companies, like Insilico Medicine and Recursion Pharmaceuticals, have also made that switch, recognizing the business value of making their own medicines: a wholly owned program will typically have a greater return on investment, should it succeed, than a licensed one.

That means that increasingly chemists will need to work in lockstep with their computational counterparts. For small molecules in particular, scientists not using AI and machine learning (ML) will be at a disadvantage to those who are, says Justin Scheer, Johnson & Johnson Innovative Medicine’s global head of in silico discovery.

But chemists may also be uniquely suited to such adaptation, says Iambic Therapeutics cofounder and CEO Tom Miller. Miller is a theoretical chemist by training who turned his attention to computational chemistry while he was a professor at Caltech. In 2022, he left the university to head up Iambic full time. The AI-driven biotech start-up is developing multiple compounds for cancer treatment.

“I think chemistry is a killer application for AI,” Miller says. “In some ways, it’s a big search problem, right? We’re searching over the possible ways to make a molecule. There’s a huge number of things you can explore. . . . It’s also a multiparameter optimization. You want your molecule to be good at this and this. That’s something that computation and AI are very good at doing: predicting across many diverse things simultaneously.”

“The future of chemistry will be AI-infused, no doubt,” he adds.

It’s also possible that chemists won’t need computational skills—they’ll just need to know the right queries to make of software tools. Nvidia’s Calad-Thomson believes that generative AI, like ChatGPT, will be a “step change” for drug discovery, making it easier for chemists “to interface with AI models without having developed that platform.” (This is central to Nvidia’s business strategy: the company offers its software as a service to drug developers.)

Chemify, a UK-based automated chemistry start-up, is already beginning this kind of upskilling. CEO Lee Cronin says that in the last 6 months, the company has trained its growing operations team to write code, “safely execute the chemistry, and get the molecules out” for multiple programs, including the manufacture of morphine.

These employees do not have advanced training in chemistry, Cronin says, and on average, it takes about 6 weeks to familiarize them with Chemify’s computational language and other systems.

“I was a bit scared about building the robots to do chemistry because people will say, ‘You’re replacing the chemist. But actually, how many postdoc chemists get up in the morning and go, ‘Yay, I want to go and spend all day on my feet in front of a fume hood, pouring flammable solvents and mutagenic materials?’” Cronin says. “We’re able to upskill people to do chemistry, and we’ve stopped de-skilling people who should be doing other things with their skills. For me, that’s a massive pat on the back for what we’re going to do culturally for chemistry.”