The AlphaFold Protein Structure Database is a resource developed through a collaboration between Google DeepMind and the EMBL European Bioinformatics Institute (EMBL-EBI), aiming to provide free access to predicted three-dimensional protein structures for the scientific community. AlphaFold is an AI system developed by Google DeepMind that can predict the three-dimensional structure of proteins from amino acid sequences, typically achieving accuracy competitive with experimental methods.
In terms of uses, the database provides predicted protein structures from humans and animal viruses with pandemic potential, which can be used to support research on viral mechanisms, vaccines, and antiviral drug development; provides predictions for bacterial proteins associated with antimicrobial resistance, supporting research on resistance mechanisms and new antimicrobial strategies; provides predicted protein structures of organisms that cause neglected tropical diseases, aiding pathogen biology and the discovery of drug or vaccine targets; it also includes predictions for the WHO's 30 priority global health proteomes, supporting research on pathogens of importance to global public health.
In terms of main capabilities, the latest version of the database contains more than 200 million entries, broadly covering UniProt (the standard library of protein sequences and annotations). The database provides separate downloads for the human proteome, the proteomes of 47 other organisms key to research and global health, and the UniProt manually curated subset (Swiss-Prot). AlphaFold was ranked as the top protein structure prediction method in CASP14, with significant advantages and high prediction accuracy. The system still has some limitations, but the CASP results indicate that AlphaFold has the immediate potential to help understand protein structures and advance biological research.
In terms of usage, users can access the database through the website to view predicted structures and metadata, with support for searching by protein or organism. The database provides download options for local analysis. If a use case is not within the database's coverage, users can use the open-source code to generate their own AlphaFold predictions, and the code also supports multimer prediction. In addition, the database integrates AI-derived residue-level annotations from scientific literature. A Transformer-based named entity recognition (NER) model identifies protein residues and point mutations in structural biology publications, and through the SIFTS framework transfers experiment-supported annotations from PDB structures to AlphaFold models, enabling users to navigate directly from residues in predictions to descriptions of their function, interactions, and experimental observations.Mutated publications.
Use cases include structural biology, drug discovery, vaccine development, antiviral and antibacterial research, pathogen biology research, and more. The database also plans to continuously update the structures of newly discovered protein sequences and improve functionality based on user feedback.
Regarding limitations, the AlphaFold system still has some limitations, but the materials do not specify them in detail. The accuracy of the database may vary by protein. Use of the data is subject to the CC-BY-4.0 license, and both academic and commercial use are permitted, but EMBL-EBI expects attribution in accordance with good scientific practice. If AlphaMissense data are used, the corresponding paper must be cited. AlphaFold data are provided by GDM, with copyright belonging to DeepMind Technologies Limited (2022); data from third-party data providers have separate copyright statements.
Citation requirements: Use of this resource requires citing Bertoni D. et al. AlphaFold Protein Structure Database 2025: a redesigned interface and updated structural coverage, Nucleic Acids Research, (2025). If AlphaMissense data are used, Cheng, J et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science (2023) must be cited.
Contact and training: If you have questions or they are not answered in the FAQ, you can send an email to alphafold@deepmind.com. EMBL-EBI provides training resources, including the recorded webinar "Accessing and interpreting predicted protein structures from AlphaFold database" and online tutorials.
