In August 2023, as part of Mark Zuckerberg’s “year of efficiency” that led to more than 20,000 layoffs, Meta disbanded a research team of a dozen scientists who had trained a large AI language model for biology.
But Alexander Rives, who led the research cohort known as Meta’s “AI protein team,” wasn’t deterred by Meta’s move. He immediately launched a startup with a core group of his former Meta colleagues, called EvolutionaryScale, to continue their work building large language models that, instead of generating text, images, or video, generate recipes for entirely new proteins.
This idea is to make biology essentially programmable, with potential applications ranging from drug development and cancer treatment (such as antibodies) to environmental protection techniques (for example, enzymes – proteins – can help break down plastic). Researchers could specify the protein’s function and other characteristics, such as its toxicity to humans, as a prompt and have the AI model return the DNA formula for producing exactly that protein.
Today, EvolutionaryScale, based in New York and San Francisco, announced it has raised more than $142 million in seed funding, led by Nat Friedman and Daniel Gross, and Lux Capital, with participation from Amazon Web Services (AWS), NVentures (Nvidia’s venture capital fund arm) and angel investors. The announcement adds Rives and his ex-Meta team to a growing list of Meta alumni – a sort of ‘Meta AI mafia’ – who have made waves with new startups in the space, most notably Mistral.
In addition to the financing, the company also announced that it had created ESM3, Rives said Fortune is a generative model for biology that has been trained on more computations than any other LLM in the space. Trained on nearly four billion proteins from the natural world, the model can simultaneously reason about the DNA sequence, physical structure, and function of proteins—three fundamental aspects of protein biology and biochemistry. And in a new paper , EvolutionaryScale showed how ESM3 was applied to generate an entirely new fluorescent protein—a type of protein first isolated in glowing jellyfish—that would have taken millions of years of evolution to create in nature.
The AI model, he explained, can process the three-dimensional structure of proteins as a language – like an alphabet of different characters – which can then be called up, just like other models, including ChatGPT. But in this case, the ‘grammar’ of proteins allows the model to use any combination of a protein’s sequence, structure and function. “We see that the model can find very creative solutions to these questions,” he said.
Training such a large-scale model requires expertise in both biology and machine learning and enormous amounts of computing power, which explains the stunning early fundraising. “They require a lot of computing power to build and train, similar to other AI frontier modeling efforts,” Rives said. The fundraising, he said, “really reflects the resources we need to do that.”
EvolutionaryScale is far from the only company focusing on the potential of generative AI-powered biology or even specifically pursuing LLMs. InstaDeep, a London-based company acquired by BioNTech in 2023, best known for its contribution to the development of the Pfizer COVID vaccine, has created an LLM in genomics, although it is not as large as the one EvolutionaryScale is working on. Profluent, a San Francisco-based AI-based biotech startup, is also focused on developing LLMs to design novel proteins. And Google DeepMind’s AlphaFold is a model that predicts the structure of proteins using a generative AI model.
Friedman, who led the funding round in EvolutionaryScale with fellow investor Gross, said Rives and the other Meta alumni are a “dream team.” (Gross recently teamed up with ex-OpenAI’s Ilya Sutskever and Daniel Levy to launch a new startup, Safe Superintelligence.)
“This was clearly the team that invented protein language modeling and had all the capabilities to scale it up,” Friedman said. “Alex thinks very big. He wants to build a fully multimodal model that encompasses all the complexities of biology. I was looking for someone with the ambition, vision, level of thinking and the expertise to make that happen.”
Rives said ESM3 will have a direct impact on scientific research, with academics able to use open versions of the model for free. The company will also offer a commercial version for pharmaceutical companies to use in drug discovery and development. This is similar to the model Google DeepMind has been pursuing, with a version of AlphaFold available for free to researchers, but with a separate spinoff company, Isomorphic Labs, working on partnerships with pharmaceutical companies.
As for Meta, Rives said he wasn’t terribly surprised when the company cut his team.
“Meta is not a biotech company,” he said. While Meta’s open research culture made it an “incredible” place to do the work, he added, “we reached the point where we really wanted to go to the next level of scaling these models. I think building a new company was really the right place to do that.”
It’s an “incredibly talented alumni group that’s come out of” Meta, Friedman said. “They’ve hired some incredible people — and I think the EvolutionaryScale team is very grateful to Meta for incubating their efforts.”