AlphaFold (Deep-Learning Protein Structure Prediction)

Deep-learning system developed by Google DeepMind that predicts 3D atomic coordinates of proteins and complexes from sequence.

Specifications & Metrics

type: AI Structural Biology & Target Modeling
language: Python (JAX / PyTorch)
license: Apache License 2.0 (AlphaFold 2 source code) and CC-BY-4.0 (AlphaFold Protein Structure Database)
developer: John Jumper, Demis Hassabis, and colleagues at DeepMind (CASP13 2018, CASP14 2020; 2024 Nobel Prize in Chemistry)

After AlphaFold 1 placed first at CASP13 in December 2018, AlphaFold 2 (2020) introduced an attention-based Evoformer architecture that scored above 90 on the CASP14 Global Distance Test (GDT), matching experimental accuracy across two-thirds of targets.
In partnership with EMBL-EBI, the AlphaFold Protein Structure Database released predicted 3D coordinates for the human proteome and over 200 million UniProt protein sequences into the public domain.
AlphaFold-Multimer and AlphaFold 3 extend structure prediction from single polypeptide chains to multi-chain protein complexes, nucleic acids, ions, and bound small-molecule ligands.