AI Breakthrough: Simulating Millions of Years of Protein Evolution in Days

A former Meta engineer’s AI research lab, EvolutionaryScale, has launched ESM3, a multimodal generative language model that can design novel proteins, simulating millions of years of evolution in a matter of days.

Groundbreaking capabilities: ESM3 is a frontier generative model for biology that can reason across the fundamental properties of proteins – sequence, structure, and function:

• The model was trained on 2.78 billion natural proteins and 771 billion unique tokens, enabling it to generate novel proteins with an unprecedented level of control.
• In tests, ESM3 generated a novel green fluorescent protein (esmGFP) that would have taken over 500 million years to evolve naturally, demonstrating its ability to accelerate protein evolution.
• ESM3 can self-improve by providing feedback on the quality of its generations, and it can be aligned with goals using feedback from lab experiments or existing data.

Potential impact and applications: EvolutionaryScale aims to make biology programmable, which could help solve major challenges like climate change, plastic pollution, and diseases:

• The technology could be particularly valuable for pharmaceutical companies developing novel medicines targeting life-threatening conditions.
• Previous models from the company have already been used to improve therapeutically relevant characteristics of antibodies and detect high-risk COVID-19 variants.
• The smallest version of ESM3 has been open-sourced to accelerate research, while larger versions are available commercially through EvolutionaryScale’s API and partners like Nvidia and AWS.

Significant funding and support: The startup has raised $142 million in a seed round led by prominent investors and tech giants, highlighting the potential and interest in this groundbreaking technology:

• The funding round was led by Nat Friedman, Daniel Gross, and Lux Capital, with participation from AWS and Nvidia’s venture capital arm.
• This substantial investment will likely fuel further development and applications of ESM3 and future models in the field of programmable biology.

Broader implications: While ESM3 represents a significant breakthrough in using AI to accelerate protein evolution and design, its real-world impact remains to be seen:

• The technology has the potential to revolutionize fields like medicine, materials science, and environmental sustainability, but it will require extensive testing, validation, and collaboration with domain experts.
• Ethical considerations and responsible development will be crucial as this powerful technology advances, given its potential to fundamentally alter the building blocks of life.
• The release of ESM3 highlights the rapid progress and expanding frontiers of AI, as models like GPT-4 and Claude continue to push boundaries in natural language understanding and reasoning.