PocketScout
Where could a drug bind? Pick a protein below (or enter your own), and this tool fetches the structure — AlphaFold's prediction from the AlphaFold Database for a UniProt accession, or an experimental structure from the RCSB PDB for a PDB id — then finds candidate binding pockets and ranks them. Everything runs in your browser: the cavities are detected live from the coordinates; nothing is installed and nothing about you is collected.
This is the tool — the core cavity finder + honest ranking. A native version, when it ships, adds a solvent-accessible-surface rendering, saved pockets, residue-level export, and offline use; it never gates the web.
A game mode: Pocket peek. Find the stretch of PTP1B that carries its active-site cysteine in four looks, with a clue after each look.
Pocket ranking
Each candidate pocket's geometric druggability score (0–100), tallest first. The score combines the pocket's volume, how enclosed it is, and how hydrophobic its lining is. Colour marks the honest verdict: well-defined, borderline, shallow.
Candidate pockets
Structure
Why the top pocket matters
In carbonic anhydrase II the deepest cleft holds a zinc ion bound by three histidines (His94, His96 and His119 in the conventional numbering, one higher in UniProt P00918). There the enzyme converts carbon dioxide and water into bicarbonate and a proton, and drugs such as acetazolamide bind to that zinc in the same cleft. A geometric score can find this cleft without knowing any of that chemistry; shape alone does not always point to a drug site (KRAS is the counter-example).
Methods & limits
- This is a geometric cavity finder, not a druggability oracle. The score is a transparent geometric heuristic — it is not a druggability predictor, not a machine-learning model, and not a docking result. A high score means a deep, enclosed, hydrophobic cavity; it does not mean a drug exists for it.
- How pockets are found: the structure is placed on a grid; free grid points enclosed by protein on multiple axes (a protein-solvent-protein / LIGSITE-style test) are kept and clustered into pockets. Volume, enclosure, and hydrophobic-lining fraction are measured per pocket.
- Single conformation, no cofactors. The finder sees one static model with waters, ions, and ligands removed — so a cavity that only opens on binding (a cryptic pocket) may be missed or shrunk, and a site built around a metal ion (e.g. a catalytic zinc) may look smaller than it functionally is.
- Predictions are models. For an AlphaFold structure, a predicted pocket is a hypothesis about a predicted fold; confidence in the fold (pLDDT) is not shown here — cross-check low-confidence regions in the viewer.
- Ranking is heuristic. The known functional site is usually near the top but not always #1 — KRAS's shallow, cryptic drug pocket is a deliberate example where geometry alone under-ranks the real target.
About this tool
Structures come from the AlphaFold Database (EMBL-EBI / Google DeepMind) and the RCSB Protein Data Bank. Both are fetched directly by your browser and rendered with Mol* via PDBe Mol*, self-hosted here — no third-party CDN, no tracking, no server. The cavity-detection approach follows the published LIGSITEcsc family of geometric pocket finders. AlphaFold DB data is released under CC BY 4.0; please cite the AlphaFold and RCSB PDB papers when you use a structure. This tool is part of the free, non-profit Fold Commons project.