AlphaFold on the platform

AlphaFold Fetch retrieves a pre-computed 3D structure by UniProt ID — a database lookup that returns in seconds with no GPU. Reach for it first when your protein is a known, catalogued sequence. If the sequence is novel, engineered, or mutated and isn't in the database, predict it with ESMFold instead; if an experimental structure has been deposited, download it from the PDB for the highest-confidence option. See what you can achieve for the full set.

Try it

What AlphaFold Fetch does

AlphaFold Fetch retrieves pre-computed protein structures from the EBI AlphaFold Database. It is a database lookup, not a prediction — the structure was computed ahead of time by DeepMind and stored as a static file. The database reports the producing pipeline in its own metadata as AlphaFold Monomer v2.0 pipeline, so what you get is an AlphaFold 2 monomer model, not AlphaFold 3 and not AlphaFold-Multimer. The database is re-released periodically rather than computed once; see translatability for what that means when you compare a model against a paper.

For proteins that are already in the database, this is the fastest way to get a structure: typically 2-5 seconds, no GPU required. It runs when you ask Ivy for it in chat; there is no form, job or CLI command for it.

Citations

Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596(7873):583-589. doi: 10.1038/s41586-021-03819-2

Varadi M, Bertoni D, Magana P, Paramval U, Pidruchna I, Radhakrishnan M, et al. AlphaFold Protein Structure Database in 2024: providing structure coverage for over 214 million protein sequences. Nucleic Acids Research. 2024;52(D1):D368-D375. doi: 10.1093/nar/gkad1011

When to fetch vs. predict

What the AlphaFold Database contains

• 98.5% of human proteins — 20,296 of the 20,614 reference-proteome entries carried a full-chain prediction in the original human-proteome release
• Major model organism proteomes (mouse, zebrafish, drosophila, etc.)
• Over 214 million predicted structures as of September 2023, up from roughly 300,000 at the July 2021 launch

If your protein is in UniProt and from a covered organism, a pre-computed structure usually exists — but coverage is not a quality claim, and translatability covers what the model is and is not good for.

Output

A standard PDB file with per-residue pLDDT confidence scores embedded in the B-factor column, plus the entry's metadata and the URLs the database publishes for it (PDB, mmCIF, binaryCIF, and the Predicted Aligned Error image and JSON).

The file is the one EBI publishes at alphafold.ebi.ac.uk/files/AF-{uniprot_id}-F1-model_v{N}.pdb. {N} is not fixed. The database deletes the old file keys when it bumps its model version, so a hardcoded version 404s for every accession once the database moves on — and that 404 reads exactly like "no model exists for this protein." The platform therefore asks the prediction API for the current pdbUrl and falls back to the pinned version only if the API is unreachable.

Workflow chaining

• Feeds into: DiffDock (docking), P2Rank (pocket prediction), PDB2PQR (protonation), MM-GBSA (complex minimization and energy report, not a binding affinity)
• Accepts from: a UniProt search (UniProt ID lookup)

Confidence interpretation

pLDDT is a per-residue, 0–100 confidence score, stored in the B-factor column of the PDB file. The four bands below are the AlphaFold Database's own, and are the bands the 3D viewer colours by. They originate with AlphaFold 2 — ESMFold borrowed the prediction head and the convention, not the other way round.

Disordered regions, signal peptides, and flexible linkers routinely score <50 — that is the model reporting absence of structure, not a defect. Use the Predicted Aligned Error (PAE) matrix for the relative placement of two domains, which pLDDT does not describe. What each band is and is not backed by, with primary citations, is in translatability.

Chat examples

• "Fetch the AlphaFold structure for P12345"
• "Get human EGFR from AlphaFold" (resolves to P00533)
• "Get the AlphaFold structure for insulin and dock aspirin against it" (chains the AlphaFold fetch → DiffDock)

Fetch, fall back, find pockets in one request

You do not need to pick tools for the standard structure pipeline. Ask for the goal ("Get the structure of P00533, or predict it if AlphaFold has none, and find its pockets") and the orchestrator chains the AlphaFold fetch, an ESMFold prediction when the fetch misses, and pocket prediction.

Read next