AlphaFold Database Download

How to download an AlphaFold structure and PAE file?

Enter a UniProt accession in ProteinIQ, such as P00533 for EGFR, then choose all available files or one output format. AlphaFold Database Download retrieves the AlphaFold DB model and available confidence data as CIF, PDB, BCIF, or PAE JSON files, ready to download and inspect alongside the source record.

What is AlphaFold Database Download?

AlphaFold Database Download is a direct retrieval tool for predicted protein structures deposited in the AlphaFold Protein Structure Database. It looks up a known UniProt accession and returns the model files that AlphaFold DB makes available for that record, rather than generating a new structure prediction.

The database is useful for starting from a published model with its confidence information intact. A downloaded model is still a prediction, not an experimental structure, and the confidence data matters as much as the coordinates when deciding whether a region or domain arrangement is suitable for downstream analysis.

Input and settings

The tool retrieves one known record at a time. If the accession is malformed, has no matching AlphaFold DB entry, or the selected file type is unavailable, no substitute model is chosen. Search AlphaFold DB first when the accession is unknown or when several proteins share a gene name.

Files returned

The files shown in the results panel preserve the selected record's source URL, format, and normalized filename. For P00533 with Output format set to All formats, the returned filenames are:

alphafold-P00533-cif.cif
alphafold-P00533-pdb.pdb
alphafold-P00533-bcif.bcif
alphafold-P00533-pae-json.json

Understanding AlphaFold DB confidence data

AlphaFold DB reports a local confidence measure, pLDDT, for each residue and a predicted aligned error, PAE, for pairs of residues. pLDDT helps judge whether the model is locally reliable at a position. PAE answers a different question: how confidently the model places one residue or region relative to another.

Reading PAE JSON

The PAE JSON contains a square predicted_aligned_error matrix. Each value estimates the positional error in angstroms for one residue when the predicted and true structures are aligned on another residue. Follow the viewer's scored-residue and aligned-residue axis labels: PAE is directional, so swapping the two residues can change the value.

Low PAE within a compact block supports confidence in the relative arrangement inside that domain. Low PAE within two domains but high PAE between them means each domain may be credible while their mutual orientation is uncertain. That distinction is important before using a full-length prediction for rigid-body docking, interpreting a long linker, or measuring a distance between domains.

PAE JSON is raw data rather than a universal structure format. It is best read with code or a viewer that explicitly supports the current AlphaFold DB PAE schema. The exported matrix is not a binding score, an experimental error estimate, or evidence that a particular conformation is populated in solution.

Choosing an AlphaFold DB file for analysis

Choose mmCIF when the next program supports it. It is the more expressive coordinate format and avoids several legacy PDB restrictions. Choose PDB for programs that only accept PDB, then open the model in PDB Viewer for a quick structural inspection. PDB to CIF can convert a PDB file when a downstream workflow requires mmCIF, although downloading the CIF directly retains the original database delivery format.

Use AlphaFold Database Download when a published AlphaFold DB model already exists for a UniProt entry. For an experimental coordinate set, retrieve the corresponding archive record with PDB Download instead. For a sequence without a suitable database model, or when a new prediction is required for a sequence variant or designed protein, run AlphaFold2 or a single-sequence predictor such as ESMFold.

Predicted structures should not be treated as automatically suitable for every task. Low-confidence residues can reflect intrinsic disorder or flexibility, and a confident isolated domain does not establish the orientation of neighboring domains. Retain the accession, model version shown on the AlphaFold DB entry, file type, and any PAE assessment with downstream results so the chosen model remains traceable.

Download availability

Enter one identifier or compound name on a single line, up to 200 characters. Surrounding spaces are ignored. These lookups need an internet connection: ProteinIQ retrieves the record from the selected database, and file downloads contact its file service.

If a text file cannot be loaded in the browser, a warning appears and its download link stays available. A listed link does not guarantee that the database currently serves that file. Retry the download or check the original database record; a connection failure is different from an unknown entry.

FAQ

Does this run AlphaFold on my sequence?

No. It retrieves an existing database model by UniProt accession. Pasted sequences, gene names, and model filenames are not accepted accession inputs.

Where are the pLDDT scores?

pLDDT is stored in the B-factor field of the downloaded PDB and mmCIF structures. It is a confidence score, not an experimental temperature factor. PAE is downloaded separately as JSON.