Pocket detection
Finds candidate binding pockets on a structure and ranks them by score and druggability, giving each a centre and a box you can dock into. Use when you have a structure and need somewhere to dock, or want to know whether a protein has a site a small molecule could occupy at all.
Pocket detection
What it does
Takes one structure and reports the cavities on its surface, best first. Each pocket comes with a score, a druggability estimate, a volume, the residues that line it, and — the part that makes it usable — a centre and a box size.
This is the step between having a structure and doing anything structure-based with it. Docking needs somewhere to dock; this is what says where.
How pockets are found
The engine is fpocket, which works by Voronoi tessellation. It fills the space around the protein with spheres that touch four atoms each ("alpha spheres"), keeps the ones small enough to sit in a cavity but large enough to hold something, and clusters what remains into pockets.
It is purely geometric. It does not know what your protein does, whether a ligand has ever been found there, or whether the cavity is real in solution rather than an artefact of one crystal form.
Ranking is fpocket's own score, which combines how buried the cavity is, how big it is, and how hydrophobic. The druggability score is separate and more specific: it estimates whether a small molecule could bind, and it is the number to look at if that is your question.
Reading the output
Three files per run:
- `_pockets.json` — every reported pocket with all its descriptors, its centre and box, and its lining residues
- `_pockets.tsv` — the same as a table you can open in a spreadsheet
- `_pockets.pdb` — the structure with the pockets marked, which opens in the 3D viewer
Druggability above 0.5 is the range where fpocket expects a small molecule could bind. Below 0.2 suggests the surface has no site worth pursuing. Between the two is genuinely uncertain and worth looking at by eye.
The box is padded by 8 Å beyond the lining atoms, because a box drawn tight to the protein leaves a ligand no room to be placed. It is in the form a docking program wants, so you can carry the numbers straight across.
Important limitations
A high score is not evidence that anything binds there. fpocket finds cavities; whether a cavity is a real binding site is a separate question that needs a ligand, a mutation, or a paper.
Crystal artefacts score too. A cavity created by how the protein packed in one crystal will be reported like any other. Comparing more than one structure of the same protein is the usual check.
"No pockets found" is a real answer, not a failure. A short peptide or a single helix has no enclosed cavity. The run is still recorded, with a note.
The count reported is capped. By default the ten best-scoring are kept; the total found is always in the summary, so a truncated list never reads as the whole answer.