PDB: 3D Structures at RCSB and PDBe

By the end of this lesson you will be able to read a Protein Data Bank record, tell which experimental method solved it, choose between the modern and legacy file formats on purpose, and pull a BRCA1 structure into Python with one call. This is where P38398 stops being a string of letters and becomes atoms with coordinates.

What the PDB actually stores

The Protein Data Bank (PDB) is the single global archive of experimentally determined 3D structures — the (x, y, z) coordinates of the atoms in a protein, nucleic acid, or complex, measured in the lab rather than predicted by software. It is a deposition (archival) database, not a curator-built one: it stores what an experimenter submitted, and it is the primary evidence that a molecule folds the way we say it does.

Every entry has a 4-character PDB ID — a short, stable accession like 1JNX or 1JM7. The first character is always a digit 1–9; the rest are alphanumeric. Because that space was running out, newer depositions also carry an extended 12-character ID such as pdb_00001jnx, but the 4-character form still works everywhere. Unlike GenBank-style accessions such as NM_007294, a PDB ID is not versioned — one ID is one deposition, and a re-refinement gets a brand-new ID rather than a .2.

One thing to internalise early: a PDB entry is almost never a whole protein. BRCA1 is 1,863 residues long, and no one has crystallised all of it. What the PDB holds are the parts that fold into stable, solvable pieces — the domains.

Reading the experimental method

The most important field in any PDB record is how the structure was measured, because it tells you what to trust. Three methods dominate:

Resolution, in ångströms (Å, one ten-billionth of a metre), is the ruler here: a lower number means atoms are placed more precisely, so a 1.3 Å X-ray structure is sharper than a 3.5 Å one. NMR entries report no resolution — they give several equally-plausible models instead, and you should treat that spread as genuine uncertainty, not noise.

BRCA1 shows all three. Its tandem BRCT domains (the C-terminal region, roughly residues 1646–1859) were solved by X-ray — 1JNX is the classic apo structure, and 4IGK catches the BRCT pair gripping a phosphorylated peptide. Its N-terminal RING domain, which pairs with the protein BARD1 to form an active complex, was first solved by NMR as 1JM7, and more recently caught by cryo-EM inside larger assemblies.

mmCIF versus legacy PDB format

Structures come in two file formats, and the choice matters more than it looks.

The legacy PDB format (.pdb) is the old fixed-column text layout: each atom on one line, its fields locked to specific character positions. It is human-readable and still everywhere, but those fixed columns cap an entry at 99,999 atoms and 62 chains — too small for a modern ribosome or a big cryo-EM complex.

mmCIF / PDBx (.cif) is the modern default, and since 2014 it has been the PDB's master format. It is a key–value layout with no column limits, room for rich metadata, and no ceiling on size. New large structures are released only as mmCIF.

The practical rule: reach for mmCIF unless a specific old tool forces legacy .pdb on you. In Biopython the parser mirrors this split.

from Bio.PDB import MMCIFParser, PDBList

# Download BRCA1's tandem-BRCT structure as mmCIF
pdbl = PDBList()
path = pdbl.retrieve_pdb_file("1JNX", file_format="mmCif")

structure = MMCIFParser(QUIET=True).get_structure("brct", path)
for model in structure:
    for chain in model:
        residues = [r for r in chain if r.id[0] == " "]
        print(chain.id, "->", len(residues), "residues")

RCSB and PDBe are mirrors

Here the two-hub story from The Two Hubs: NCBI Entrez and EMBL-EBI repeats, one level up. The PDB is run by the wwPDB (worldwide PDB) partnership, and its members serve the same archive through different portals:

• RCSB PDB (rcsb.org), the US site.
• PDBe (ebi.ac.uk/pdbe), the European site at EMBL-EBI.
• PDBj in Japan, and BMRB for NMR data.

Deposit at any one and the entry propagates to all — so 1JNX is byte-for-byte the same structure at RCSB and at PDBe. What differs is the tooling wrapped around it: RCSB's search and its data API, PDBe's validation reports and its residue-level annotations. Pick the portal whose extras you want; the coordinates underneath are identical.

You can fetch either portal's data straight from a URL. RCSB serves per-entry files at https://files.rcsb.org/download/1JNX.cif, and both sites also expose REST APIs for scripted retrieval.

Experimental versus predicted

One last boundary. Everything in the PDB was measured. That is its whole value — and its limit, because most proteins have never been crystallised. For the gaps, structure prediction (AlphaFold and its successors) now supplies computed models, and the UniProt entry P38398 links out to both: solved PDB fragments and a full-length predicted model side by side. The two are not interchangeable — a prediction is a hypothesis, a PDB entry is evidence — and our blog on the AlphaFold revolution walks through exactly when to trust which. Keep the line clear, and the PDB will keep telling you what is real.