FreeHow credits work →
AF
Predicted structuresAlphaFold DB prediction API

AlphaFold DB

The AlphaFold Protein Structure Database holds predicted 3D protein structures and is run by Google DeepMind and EMBL-EBI. Through ScholarMuse, an agent can fetch the predicted model for a UniProt accession and read its confidence scores and file links. The model is returned next to the experimental PDB entries for the same protein, so the two can be compared in one call.

ScholarMuse tools: get_structure. Each successful call uses credits.

Ask your agent

Is there an AlphaFold model for UniProt P00533, and what is its mean pLDDT?
How much of the AlphaFold prediction for human TP53 is low or very low confidence?
Give me the mmCIF link for the AlphaFold model of the protein in PDB entry 1M17.

What your agent gets back

Every record carries id, source, title, url, license, attribution and retrievedAt. The fields below are the data that comes from AlphaFold DB.

get_structure

One record with kind "predicted", only when a UniProt accession is known and AlphaFold DB has a model for it.

entryId
AlphaFold DB entry id, e.g. AF-P00533-F1
uniprotId
UniProt accession of the model
gene
Gene name
organism
Organism scientific name
meanPlddt
Mean pLDDT confidence score
fractionVeryHighConfidence
Fraction of residues with very high pLDDT
fractionLowConfidence
Fraction of residues with low pLDDT
version
Latest model version
modelDate
Date the model was created
residueStart
First UniProt residue covered
residueEnd
Last UniProt residue covered
mmcifUrl
mmCIF file download link
paeImageUrl
Predicted aligned error image link

What it covers

  • Predicted protein structures in AlphaFold DB, looked up one UniProt accession at a time.
  • One model per call: the one for the canonical sequence. The isoformModels field counts the other models AlphaFold DB returned.
  • When you pass a PDB id, the model returned is for the first UniProt accession listed in that entry.

Limits

  • AlphaFold DB cannot be searched by name here. A name works only when it exactly matches a human gene symbol or protein name in UniProt; otherwise no predicted model is returned.
  • Returns confidence summaries and file links, not coordinates or per-residue pLDDT values.
  • Proteins without a model in AlphaFold DB return nothing from this source.

How ScholarMuse reaches AlphaFold DB

The connector calls the AlphaFold DB API at alphafold.ebi.ac.uk/api/prediction/{UniProt accession}. No API key is sent.

Responses are cached for up to 7 days, so a new model version can take that long to appear.

API key: the upstream API can be called without one, so you do not add one.

Licence and attribution

Licence: CC BY 4.0

Predicted structure from AlphaFold Protein Structure Database (alphafold.ebi.ac.uk), Google DeepMind and EMBL-EBI, CC BY 4.0.

Records that come from AlphaFold DB carry this licence and attribution. When you cite the data in a paper, follow AlphaFold DB's own citation guidance.

Related sources

FAQ

ScholarMuse is not an official AlphaFold DB product. It is one MCP server for academic data, and its get_structure tool queries AlphaFold DB. An agent connected to ScholarMuse can use AlphaFold DB without a separate server.

No. The AlphaFold DB API can be called without a key, and ScholarMuse does not ask you for one. You sign in to ScholarMuse once and your agent calls the tool.

The connector calls the AlphaFold DB API at alphafold.ebi.ac.uk/api/prediction/{UniProt accession}. No API key is sent. Responses are cached for up to 7 days, so a new model version can take that long to appear.

CC BY 4.0. Attribution: Predicted structure from AlphaFold Protein Structure Database (alphafold.ebi.ac.uk), Google DeepMind and EMBL-EBI, CC BY 4.0. Records that come from AlphaFold DB carry this licence and attribution.

Yes. Each successful call to get_structure uses credits, and a call that fails is not charged. Current price: Protein structures, 1 credit / call.