Protein Structures — AlphaFold & PDB

Protein Structures — AlphaFold & PDB

How to get a protein's 3D structure and show it to the user. Most "predict the structure of protein X" requests do not require running AlphaFold — the AlphaFold DB already holds a precomputed prediction for nearly every known protein. Run a prediction only for a sequence that is not a known UniProt entry.

1. AlphaFold DB — predicted structures (the usual path)

The AlphaFold Database has precomputed AlphaFold models for ~200M+ UniProt proteins. If the protein is a known UniProt entry, its predicted structure already exists — just fetch it (instant, free).

Use the API — do not hand-build file URLs (the model version, currently v6, changes; the API always returns the live URLs):

import requests

acc = "P00533"  # UniProt accession (human EGFR)
meta = requests.get(
    f"https://alphafold.ebi.ac.uk/api/prediction/{acc}", timeout=30
).json()[0]
cif_url = meta["cifUrl"]          # also: pdbUrl, bcifUrl
with open(f"{acc}.cif", "wb") as fh:
    fh.write(requests.get(cif_url, timeout=60).content)

No UniProt accession yet? Resolve a gene name / protein name to an accession first via the UniProt REST API (https://rest.uniprot.org/uniprotkb/search?query=<gene>+AND+organism_id:9606&format=json) or the gget package — pick the reviewed (Swiss-Prot) entry.

2. RCSB PDB — experimental structures

For a solved, experimental structure, download from the RCSB PDB by id:

https://files.rcsb.org/download/<PDBID>.cif      (or .pdb)

3. Predicting a novel sequence

If the sequence is not a known UniProt protein (a designed, mutant, or synthetic sequence), the AlphaFold DB has nothing — you must predict it.

ESMFold API — the practical path (no GPU)

ESMFold (Meta's structure predictor) has a public folding API: POST a raw amino-acid sequence, get a PDB structure back. No GPU, no install, no databases — this is the realistic way to predict a novel sequence here.

import requests

seq = "MALWMRLLPLLALLALWGPDPAAA..."          # one-letter, a single chain
resp = requests.post("https://api.esmatlas.com/foldSequence/v1/pdb/",
                     data=seq, timeout=180)
resp.raise_for_status()
with open("predicted.pdb", "w") as fh:
    fh.write(resp.text)