AlphaFold DB

AlphaFold DB

AlphaFold DB is EMBL-EBI's public archive of AlphaFold-predicted protein structures; this connector searches it directly from inside RxCubed. It's available under Collect files. Search the way you would UniProt — protein name, gene, accession, or organism, such as “carbonic anhydrase 2 human” — because AlphaFold models are keyed to UniProt accessions. See File Collection for the shared search, shortlist and save flow every connector here uses.

What to compare

The result card shows accession, subtitle, organism, gene, model version, sequence coverage, and mean pLDDT.

pLDDT is the number that matters. It is AlphaFold's own confidence in each residue's position, averaged here across the model:

• above 90 — high confidence, backbone and side chains usually reliable;
• 70 to 90 — confident backbone, less certain side chains;
• 50 to 70 — low confidence, treat with caution;
• below 50 — usually disordered rather than wrong; do not read structure into it.

AlphaFold's own result order is shown; this connector does not use AI scoring, so there are no Qualities or importance controls.

Files

• Coordinates (mmCIF) — the predicted model, selected by default.
• Coordinates (legacy PDB) — the same model in the older format, for tools that need it.
• PAE matrix (JSON) — predicted aligned error, which says how confident the model is about the relative position of any two parts. Use it before trusting a domain arrangement.
• Confidence (JSON) — the per-residue pLDDT values behind the mean shown on the card.

A prediction is not an experiment

An AlphaFold model is a hypothesis about shape. It is not evidence that a protein adopts that shape in your conditions, and it carries no ligands, no ions, no waters, and no alternative conformations. Where an experimental structure exists, prefer it — collect it from RCSB PDB. Use AlphaFold when nothing experimental exists, and say so in any report.

What to do next

Open a saved coordinate file from Results Hub to inspect it in the 3D viewer. Predicted models are reasonable inputs to pocket detection and structure alignment, keeping the confidence caveat in mind.