RCSB Protein Data Bank

The RCSB Protein Data Bank is the US access point to the single global archive of experimentally determined three-dimensional structures of proteins, nucleic acids and complex assemblies. The underlying archive is curated by the Worldwide Protein Data Bank (wwPDB) partnership, and RCSB PDB (run from Rutgers and UC San Diego) adds the search, annotation and delivery layer at rcsb.org. It holds over 250,000 experimental structures solved by X-ray crystallography, cryo-electron microscopy and NMR, and now serves these next to more than one million computed structure models drawn from the AlphaFold Database and ModelArchive, clearly flagged so you never confuse a prediction with an experiment.

You access it in several ways. Individual coordinate and metadata files come from files.rcsb.org in PDBx/mmCIF (the primary format going forward), legacy PDB and PDBML/XML. A REST and GraphQL data API at data.rcsb.org returns structured JSON for programmatic queries, and the whole archive can be mirrored by rsync for bulk work (FTP has been retired). Each entry carries rich machine-readable annotation: sequence, ligands, resolution, experimental method and publication.

For RAG and AI, the PDB is less a text corpus than a structured knowledge base: it grounds answers about what a biomolecule looks like, which ligands bind it and how a structure was determined. A practical approach is to turn each entry's metadata into a short natural-language summary before embedding, and to resolve PDB identifiers rather than indexing raw coordinate files. It is best for drug-discovery assistants, structural-biology tooling and any system that needs authoritative structure facts.

All PDB data files are released under CC0 1.0, a public-domain dedication, so you may reuse them freely, including commercially, with no share-alike or attribution obligation (citing the original depositors is encouraged as courtesy, and attribution can apply to some integrated external resources). It pairs naturally with our UniProt entry for protein sequence and function, and sits upstream of the predicted structures in AlphaFold DB.