How to dock against an AlphaFold structure (any protein, no crystal needed)

Most proteins have never been crystallized. If your target has no PDB entry, classical docking seems out of reach — but AlphaFold has predicted structures for essentially the whole of UniProt (200M+ models). Here's how to dock a ligand against an AlphaFold structure in the browser, and — just as important — when to trust the result.

Step 1 — Import the AlphaFold model

On the Receptors page, use the AlphaFold box. You can enter either a UniProt accession (e.g. P00533 for EGFR) or simply a gene/protein name (EGFR, BRAF, KRAS). MolHub resolves the accession (reviewed/SwissProt, human-first), downloads the current AlphaFold model, and turns it into a dockable receptor. FPocket then finds cavities and suggests a docking box.

Step 2 — Dock as usual

From there it's the normal flow: pick a ligand from the library, paste a SMILES, or run a batch screen of a dataset. See the 5-minute docking walkthrough for the full steps. AutoDock Vina runs on CPU and the pose viewer highlights contact residues.

The caveats — read this before you over-trust it

AlphaFold models are predictions, and a few things matter:

• Confidence varies along the chain. AlphaFold reports per-residue pLDDT. Well-folded domains are usually reliable; flexible loops and disordered regions are not. If your pocket sits in a low-confidence region, be skeptical.
• Models are apo (ligand-free). The predicted pocket may not reflect the induced-fit conformation a ligand actually binds, so poses can be off.
• Side-chain rotamers are approximate. For tight pockets this can change the docking box and the top pose meaningfully.

Net: AlphaFold docking is excellent for generating hypotheses and triaging libraries against targets you otherwise couldn't touch. It is not a substitute for an experimental structure when one exists — if there's a good crystal in the PDB, use it.

A practical workflow

• Import the AlphaFold model by gene name.
• Check the pocket sits in a high-confidence region of the model.
• Batch-screen a focused library and rank by affinity.
• For top hits, inspect the pose + contact residues and read ADMET flags.
• Confirm experimentally — docking is enrichment, not proof.