SaProt
modelFajie Yuan's AI4Protein lab's anchor protein language model: a structure-aware PLM that combines the amino-acid sequence with a Foldseek 3Di "structural alphabet" over ~40M sequences/structures (ICLR 2024 spotlight, top-5%). SaProt ranked #1 on ProteinGym, with 100k+ HuggingFace downloads on the 650M checkpoint, and was extended in 2025 by SaProt_1.3B_AFDB_OMG_NCBI — the seed of the lab's broader protein-FM stack (ProTrek, SaprotHub, Evolla).
Model Details
Architecture DENSE
Variants
| Name | Parameters | Notes |
|---|---|---|
| SaProt-650M | 650M | — |
| SaProt-1.3B | 1.3B | — |