EvolutionaryScale, an AI startup backed by Amazon and NVIDIA, is using large language models (LLMs) to generate new proteins, potentially advancing fields like medicine and evolutionary biology.

Their third model, ESM3, successfully created a new green fluorescent protein (GFP) by simulating 500 million years of evolution.

ESM3, which is open for use by the scientific community, is trained on 771 billion tokens from 3.15 billion protein sequences.

This achievement supports the idea of determinism in evolution, as the artificial protein’s genetic sequence closely mirrors its natural counterpart.

The model has wide applications, including creating synthetic proteins for environmental solutions, like breaking down plastic waste, Popular Mechanics has reported.

EvolutionaryScale aims to empower humanity to solve global challenges with AI-driven biological innovations.

Written by B.C. Begley