RCSB Protein Data Bank

The RCSB Protein Data Bank is the US member of the worldwide PDB consortium and the single archive structural biology deposits into. It holds roughly 258,616 entries: about 258,222 experimentally determined structures - solved by X-ray diffraction, solution NMR or electron microscopy - plus 394 integrative or hybrid models that combine techniques. Alongside them sit more than one million Computed Structure Models, roughly 992,732 imported from the AlphaFold DB and 69,326 from ModelArchive, held in a separate tier so predicted coordinates never pass silently as measured ones.

Depth travels with every entry: determination method, resolution in angstroms, polymer sequences, source-organism taxonomy, keyword classification, the primary citation, and the record's own deposit, release and revision timestamps. Within the corpus, 82,833 structures cover human sequences and 22,212 entries contain nucleic acids. Get a sample of this dataset and inspect real records before committing pipeline time.

One record per entry, with identity, measurement verdicts and provenance side by side. Entry 4HHB - human deoxyhemoglobin, the structure the field calibrates against:

rcsb_id                  4HHB
struct.title             THE CRYSTAL STRUCTURE OF HUMAN DEOXYHAEMOGLOBIN
                         AT 1.74 ANGSTROMS RESOLUTION
exptl.method             X-RAY DIFFRACTION
resolution_combined      1.74
deposit_date             1984-03-07
initial_release_date     1984-07-17
status_code              REL

The same identifier space supports ranked similarity lookups, which turn one structure into a family. A sequence-similarity query against 4HHB's polymer entity returns scored neighbors:

identifier    score
3GOU          1.00000
6IHX          0.99957
2PGH          0.99863

Read together, the two blocks cover both halves of a structure workflow: the flat record stating what was measured, how well, and when it entered the archive; and the ranked neighborhood around it. Hanging off that flat row sit the deeper layers - polymer-entity records, biological assemblies, ligand geometry, and the atomic coordinates themselves.

Nineteen documented fields travel on every entry record, covering identity, experimental verdicts, composition, sequence, taxonomy, citation and lifecycle state. Definitions below follow the documented schema; entries flagged in the footnote are folded into your sample on request.

Geography - the subject is global science rather than any jurisdiction: depositions arrive from laboratories worldwide, and the organism axis spans thousands of taxa. 82,833 entries cover human sequences while the rest spread across model organisms, pathogens and plant life. Taxonomy is tagged at the polymer-entity level (rcsb_entity_source_organism), so an organism-scoped subset is a filter, not a rebuild.

Temporal - the archive runs continuously since 1971, making it the longest-lived structured collection in the life sciences. Time slicing happens per entry through rcsb_accession_info fields rather than by snapshot version: 4HHB was deposited 1984-03-07 and released 1984-07-17, and both facts ride on the record itself. One scheduled transition matters downstream - from July 21, 2027 new identifiers extend to 12 characters and mmCIF becomes the sole coordinate format, so any parser keyed to legacy four-character IDs or legacy formatting needs a migration path before then.

Granularity - four levels deep: whole entry, polymer entity, biological assembly, and individual atom coordinates, with non-polymer (ligand and cofactor) entities recorded beside the chains. A question like 'every kinase structure below 2.0 angstroms solved by electron microscopy' is a filter over indexed attributes such as exptl.method and rcsb_entry_info.resolution_combined, not a literature hunt.

Scale check: roughly 258,616 archived entries, 82,833 covering human sequences, 22,212 containing nucleic acids, and more than one million Computed Structure Models layered above. Forty-plus years of solved structures cannot be regenerated by any single lab or consortium - the corpus compounds, which is why it has no practical substitute.

API, files, or your warehouse. Daily, weekly, or hourly.

• Drug design and binding-site work - pull every structure of a target or its homologs with rcsb_entry_info.resolution_combined as the quality gate, then read ligand geometry from the non-polymer entities to compare how different chemotypes occupy the same pocket.
• Structural ML training sets - assemble labeled corpora at scale while keeping the tiers honest: experimental coordinates for ground truth, the Computed Structure Models tier for augmentation, never mixed silently.
• Target-crowding scans for competitive intel - structure counts per protein family are a public proxy for how heavily a mechanism has been worked; struct_keywords.pdbx_keywords and organism tags make the rollup a group-by.
• Enzyme engineering - sequence-similarity neighborhoods surface homologs of a scaffold across organisms, giving protein designers a spread of backbones and stability behavior to recombine.
• Nucleic-acid and complex research - 22,212 entries containing DNA, RNA or hybrids support ribosome, spliceosome and gene-editing questions at the assembly level rather than chain-by-chain.
• Citation-grade reference - every entry carries rcsb_primary_citation with journal, year and DOI, so claims made in reports and articles resolve to the original depositor's paper.

Developers and data-product builders get a stable identifier space (rcsb_id) that keys deterministically into entity, assembly and coordinate layers - the rare corpus where a foreign key still resolves after decades. Data scientists and ML engineers get numeric resolution values, method enums and composition counts as ready-made filters, plus a clean experimental-versus-computed split for training-set hygiene. Market researchers and consultants read structure density per protein family as an activity signal before sizing a therapeutic market. Investors and diligence teams treat deposition volume around a named target as evidence a mechanism is crowded or overlooked. Journalists, academics and students get citable, image-ready answers to 'what does this molecule actually look like', backed by DOIs on every record.

Notes that pair well with this page:

• Biotechnology data hub - the pooled industry view this record sits inside, alongside sequence, expression and bioactivity corpora.
• The RCSB PDB source profile - archive scope, field-dictionary layout and governance of the underlying resource.
• RCSB PDB vs PubChem - macromolecular geometry against flat small-molecule records; the pair brackets where structure ends and chemistry enumeration begins.
• Human Protein Atlas structures note - experimental coordinates beside atlas-published predicted models for the same targets.
• UniProt protein sequence data - the functional annotation layer these structures cross-reference on sequences.
• STRING protein interaction data - put each solved structure back into its interaction context.