ESM2

Meta AI's transformer-based language model that predicts protein structure and function directly from amino acid sequences.

ESM2 represents the state of the art in protein language modeling, trained on 250 million protein sequences to achieve atomic-level resolution. By leveraging a transformer architecture with up to 15 billion parameters, it outperforms previous iterations like ESM-1b in predicting secondary structures and long-range contacts. Researchers use ESM2 to accelerate drug discovery and protein engineering, as it generates high-accuracy 3D structures (ESMFold) orders of magnitude faster than AlphaFold2.

https://github.com/facebookresearch/esm