Structures · Database
AlphaFold DB
A database by Google DeepMind and EMBL-EBI providing over 200 million pre-computed protein structure predictions.
What AlphaFold DB is
AlphaFold Protein Structure Database (AlphaFold DB) is an open-access repository created through a collaboration between Google DeepMind and the European Molecular Biology Laboratory's European Bioinformatics Institute (EMBL-EBI). It provides pre-computed three-dimensional models of proteins predicted by the AlphaFold artificial intelligence system directly from amino acid sequences.
Experimentally determining three-dimensional protein structures through techniques like X-ray crystallography, cryo-electron microscopy, and nuclear magnetic resonance is resource-intensive and time-consuming. AlphaFold DB addresses this bottleneck by providing structural predictions at scale, offering broad coverage across UniProt and Swiss-Prot, including complete proteomes for humans and 47 other key research and global health organisms.
The resource maintains curated focus datasets for pandemic preparedness, antimicrobial resistance, neglected tropical diseases, and priority pathogens identified by the World Health Organization. Recent additions incorporate literature-derived, residue-level functional annotations and point mutation data linked directly to structural coordinates.
How it is organised
AlphaFold DB hosts structures predicted by AlphaFold, an end-to-end deep learning system. The underlying neural network model iteratively refines residue-residue relationships and multiple sequence alignments using an attention mechanism before passing data to a structure prediction module. Final predicted atomic coordinates undergo local physical refinement via energy minimisation based on the AMBER force field.
The database also integrates biological context through natural language processing. A transformer-based named entity recognition model scans structural biology publications to extract mentions of residues and point mutations. Using the SIFTS framework, experimentally supported annotations from Protein Data Bank entries are mapped directly onto the corresponding residues of AlphaFold models.
Key facts
- Database category
- Protein structure database
- Structure count
- Over 200 million predicted structures
- Proteome downloads
- Human proteome and 47 other key organisms
- Primary sequence reference
- UniProt and Swiss-Prot
- Refinement method
- AMBER force field energy minimisation
What you can use it for
- Antimicrobial resistance research
- Inspect predicted bacterial protein structures linked to drug resistance to investigate resistance mechanisms and identify potential antimicrobial targets.
- Pandemic preparedness and virology
- Analyse structural models of viral proteins from human and animal viruses of pandemic potential to support antiviral discovery and vaccine development.
- Neglected tropical disease pathogen biology
- Investigate proteins from pathogens causing neglected tropical diseases to uncover biological pathways and identify candidates for therapeutic intervention.
- Literature-linked residue and variant interpretation
- Explore residue-level functional evidence and point mutations extracted from publications to evaluate how specific sites affect protein structure and function.
Query it here
The research agent queries AlphaFold DB directly — no account, no API key — and every answer links back to the record it came from. Ask in plain language and it writes the query.
Strengths and limits
Good at
- Provides broad structural coverage across the UniProt database, spanning hundreds of millions of protein sequences.
- Offers pre-packaged, full-proteome downloads for humans alongside 47 other key model and global health organisms.
- Achieves prediction quality that regularly reaches levels competitive with experimental structure determination methods.
- Integrates text-mined residue annotations and point mutations linked directly to scientific literature via the SIFTS framework.
Watch out for
- Predictions do not resolve the dynamic folding pathways or the underlying physical mechanisms by which proteins fold in nature.
- A portion of predictions remain below high-accuracy thresholds, with roughly one-third scoring lower on standard distance metrics during benchmark evaluations.
- The database consists of pre-computed monomer structures, requiring users with novel sequences or multimeric interaction queries to run prediction pipelines externally.
Questions
Can I search AlphaFold DB without writing code?
Yes. The research agent queries AlphaFold DB directly, with no account or API key, and cites the record behind every answer. Ask in plain language — for a gene, a protein, a compound or a study — and it builds the query for you.
Can I reuse data from AlphaFold DB?
CC BY 4.0. Terms: https://alphafold.ebi.ac.uk/faq
What sequence sources does AlphaFold DB cover?
AlphaFold DB provides broad coverage of the UniProt database, including the manually curated Swiss-Prot subset, complete proteomes for humans and 47 other key organisms, and targeted datasets for viral, bacterial, and neglected disease pathogens.
How are residue annotations linked to the predicted structures?
A transformer-based named entity recognition model identifies residues and mutations in published papers. These are transferred from experimental Protein Data Bank structures to AlphaFold models using the SIFTS framework.
Can I query protein complexes or multimers in AlphaFold DB?
The database primarily hosts pre-computed monomeric predictions. For multimeric complexes or sequences not present in the repository, predictions must be generated using AlphaFold's standalone open-source code.
How to cite it
Bertoni D. et al. AlphaFold Protein Structure Database 2025: a redesigned interface and updated structural coverage, Nucleic Acids Research, (2025)
Written from AlphaFold DB's own website and reference material. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.