Skip to content

Meta FAIR

Catalog entries using this tag (links open the entry card on its page):

Entries

ESMFold

AI Protein Language Model ESM Meta FAIR Structure Prediction
PUBMED_LINK
36927031
FULL NAME
ESMFold — Evolutionary-Scale Protein Structure Prediction with a Language Model
DESCRIPTION
ESMFold from Meta FAIR uses a protein language model (ESM-2) trained on 65 million protein sequences via masked language modeling to predict protein 3D structures directly from sequence, without requiring multiple sequence alignments (MSAs). 60-100x faster than AlphaFold2 while maintaining high accuracy. Enables structure prediction at evolutionary scale — 617 million predicted structures released. Represents a paradigm shift combining protein language models with structure prediction.
URL
https://github.com/facebookresearch/esm
TITLE
Evolutionary-scale prediction of atomic-level protein structure with a language model.
Main citation
Lin Z, Akin H, Rao R, Hie B, Zhu Z, Lu W, Smetanin N, Verkuil R, Kabeli O, Shmueli Y, dos Santos Costa A, Fazel-Zarandi M, Sercu T, Candido S, Rives A. (2023) Evolutionary-scale prediction of atomic-level protein structure with a language model. Science, 379(6637):1123-1130. doi:10.1126/science.ade2574. PMID 36927031
ABSTRACT
Protein language models can learn evolutionary patterns from sequences without explicit alignment or structural data. Here we demonstrate that direct inference of structure at scale is possible by training a language model, ESM-2, on 65 million protein sequences and using it to predict structure. ESMFold predicts structure 60x faster than AlphaFold2 while maintaining high accuracy, enabling evolutionary-scale structural biology. We release 617 million predicted protein structures, covering the majority of sequences in the UniRef50 database.
DOI
10.1126/science.ade2574