Overview
EvolutionaryScale's core product, ESM3, is a family of generative AI models designed for protein sequence modeling. It leverages a vast dataset of 2.78 billion natural proteins and 771 billion unique tokens to enable scientists to understand, imagine, and create proteins with emergent reasoning capabilities. ESM3 models can simultaneously reason over sequence, structure, and function, allowing users to input mixed data types and explore diverse possibilities. The models are available in small, medium, and large sizes through an API, with the ESM3-open model offered with weights and source code under a non-commercial license. Use cases include designing novel proteins, enzymes for plastic breakdown, and new medicines, using chain-of-thought prompting to evolve proteins beyond natural limits, like the esmGFP, a vast departure from naturally occurring fluorescent proteins. ESM3's architecture supports integration via Forge, AWS Sagemaker, Omics platform, AWS Bedrock, BioNemo, and NVIDIA NIM microservices, enabling deployment across varied scientific environments.
