ESM is a family of large protein language models trained on evolutionary-scale sequence data. Variants (ESM-1b/2/ESMFold/ESM-3) support representation learning, zero-shot mutational effects, and structure prediction.
Type
Protein language model
Primary tasks
Embeddings, structure prediction (ESMFold), design
Input
Amino acid sequences (e.g., FASTA)
Origin
Meta AI
License
MIT
Resources
- GitHub: facebookresearch/esm
- Paper (ESMFold): Protein structure prediction using ESM
- Model card(s): see repo
Use ESM embeddings for downstream tasks (e.g., fitness prediction) or ESMFold for fast single-sequence structure prediction.