EvolutionaryScale came out of stealth today, unveiling ESM3 - a powerful new AI model that designs proteins by simulating millions of years of evolution in moments.
The New York startup, founded by former Meta AI researchers, also announced $142 million in seed funding. Tech investors Nat Friedman, Daniel Gross, and Lux Capital led the round, with Amazon and NVIDIA's venture arm also participating.
ESM3 is a language model for biology. It can reason about a protein's sequence, structure, and function all at once. The team trained it on 2.78 billion proteins from across nature.
"ESM3 lets us engineer biology from first principles, like we engineer machines or write computer programs," said Alexander Rives, EvolutionaryScale's co-founder and chief scientist.
To show off ESM3's abilities, the team created a new green fluorescent protein (GFP). GFPs occur naturally in some sea creatures and are crucial tools in biology research. The new GFP shares only 58% of its sequence with its closest natural relative. EvolutionaryScale says this much change would take over 500 million years to happen in nature.
This ability to quickly generate new, working proteins could be huge for science. It might speed up cancer drug development or help create enzymes to break down pollutants.
EvolutionaryScale is also focusing on responsible innovation. They've outlined plans for transparent development, risk assessment, and working with government and policy experts.
The company is releasing an open version of ESM3 for researchers. They're also partnering with Amazon Web Services and NVIDIA to make the full model available to scientists worldwide.
AWS will make the full ESM3 model family accessible to hundreds of thousands of researchers and nine out of the ten largest global pharma companies through its AI services. This will allow researchers to fine-tune ESM3 models using their own data securely and at scale. NVIDIA also shared that they are optimizing all versions of ESM3 for better training and inference performance. The model will be available through NVIDIA's BioNeMo platform and supported by NVIDIA AI Enterprise software.
EvolutionaryScale joins a growing field of AI startups in biotech. But with its unique approach to protein design and major funding, it's well-positioned to make waves.
ESM3 could speed up drug discovery, improve biofuels, and drive new materials science. By letting scientists "program" biology this way, EvolutionaryScale might change how fast we can make scientific breakthroughs. Read the preprint of their paper here.