Generate Biomedicines AI in healthcare

What would it take to persuade you to quit a secure job to join a start-up? For Michael Nally, chief executive officer of Generate: Biomedicines, a meeting with a Nobel laureate did the trick.

The encounter in question was with Frances Arnold, who was awarded the Nobel Prize in Chemistry in 2018 for conducting the first directed evolution of enzymes – creating more deliberate and targeted catalysts for chemical reactions.

“Mike, nature has only sampled one drop of water in all the Earth’s oceans of potential proteins,” Arnold told Nally. “And you now have a technology that can survey the oceans.”

At the time, this technology was most commonly referred to as machine learning, but it has now evolved into what is called generative artificial intelligence (Gen AI). Arnold’s contention was that such immense computing power could be applied to the process of associating DNA sequences with the function of proteins. Proteins are the molecular links of life, but we still don’t fully understand the biological function of every chain they form. Investigating them has traditionally relied on extremely labour-intensive research that aims to translate knowledge of these relationships into novel therapeutic programmes.

Nally grasped the potential, and it was enough to convince him to leave his leadership role at a large multinational pharmaceutical company with 72,000 employees to join Generate, then a 30-person start-up near Boston, in 2021.

“Many people thought I was crazy,” he recalls. “But what was troubling me when I looked across the entirety of the biopharma ecosystem was that the industry was fundamentally struggling with two big challenges. One was that research productivity had been going down for four decades in a row. And the second was that the pricing of medicines led to inadequate access, ultimately undermining the reputation of the industry. I believed that in order to address these challenges, you would ultimately need a revolution in research productivity.”

Drug development: from theory to reality

This is no longer just theory. The company has identified a monoclonal antibody (a lab-made protein that can target specific bacteria or viruses) that is already in a phase I trial for a spike protein in SARS-CoV‑2, and has successfully dosed the first patient in a study of another monoclonal antibody for the treatment of mild-to-moderate asthma.

“This technology will change the way drugs are made,” Nally argues. “It will transcend the human condition in a number of different disease areas. What excites me the most is seeing the impact, from the theoretical to the real. We’ve just seen our early phase I results, and they’re extraordinary. This technology is going to give us a lot of answers for some of the worst conditions that face humanity.”

Much more is expected. For example, the company’s AI platform can generate molecules entirely through computing power without a template in order to hit historically hard-to-drug or undruggable targets – the biological structures in the body with which a drug interacts to produce an effect – with great accuracy. This process of producing and validating “de novo” antibodies has now been demonstrated across nine distinct targets.

Chroma browsers

A fascinating aspect of Generate’s research is that last year the company published full details of Chroma, its generative AI model, in Nature. It also released the code behind Chroma as open-source software. Nally admits it was a counterintuitive move for a life sciences business, where patents have long been cherished.

“One of the big challenges we were facing was how to become the destination of choice for unicorn talents,” he explains. “We thought the best way to do that was to show how we were embracing these cutting-edge techniques.” After publishing Generate’s Chroma manuscripts, Nally recounts, the company had five machine learning internships available the next summer.

“We had 2,000 applicants for those spots. People could actually readily look and see that the capabilities and the people that we were attracting were among the best in the world.

"These are the people that they wanted to be close to in order to learn and grow from. We can’t win talent versus mega-cap tech companies if we’re in a dollar-for-dollar arms race. What we can do is provide people with a distinct sense of purpose. The fact that their work will not add to website clicks but will actually be around how you solve and cure cancer drives a certain type of scientist to a company like Generate.”

Attracting, embracing and training this next generation of scientists could propel research to even greater heights. “When I joined Generate, there were 90 people on the planet skilled at the intersection of protein engineering and machine learning,” remarks Nally. “We had historically trained people in these different silos. We’ve never broken through these silos and said, ‘It’s actually the intersection of these disciplines where the magic happens.’”

This prospective combination of human intelligence and artificial intelligence is compelling, given the scale made possible by the latter.

Nally observes that, when Generate launched, the biggest experiment they could undertake would involve around 100 data points on an individual target. “Now we routinely do datasets of a million defined variants to an individual target, and we can do them in about a quarter of the time it would normally take historically,” he states.

For all the optimism, it is of course far too soon even to think of curing all diseases as a phenomenon.

“Whenever I hear people say, ‘We’re going to solve biology purely computationally,’ I kind of laugh at that concept,” Nally says. “We just don’t have the data to have the answers that we ultimately need.” But the rapid progress made so far warrants a little dreaming about what the future holds.