Skip to main content
Documentation

Pocket discovery

Pocket discovery is the process of identifying regions on a protein’s surface where small molecules (drugs) are most likely to bind. OpenDDE uses P2Rank, a machine-learning tool developed at the Czech Technical University, to predict these binding pockets.

What is a binding pocket?

Proteins are large, complex 3D molecules. Their surfaces have grooves, clefts, and cavities. A binding pocket is a specific cavity where a drug molecule can physically fit and form chemical interactions (hydrogen bonds, hydrophobic contacts, salt bridges).

Think of it like a lock: the pocket is the keyhole, and the drug is the key. For a drug to work, it must fit snugly into the right pocket on the right protein.

🔬
Protein surface
Complex 3D topology
🕳️
Binding pocket
Cavity with chemical features
💊
Drug molecule
Fits into the pocket

How P2Rank works

P2Rank is a machine-learning method that predicts ligand-binding sites from a protein structure. Here’s how it works:

  1. Surface point sampling — The protein surface is sampled as a set of points using a Connolly surface algorithm
  2. Feature extraction — For each point, P2Rank computes chemical and geometric features: hydrophobicity, charge, surface curvature, atom density, etc.
  3. ML scoring — A random forest classifier scores each point for its likelihood of being part of a binding site
  4. Clustering — High-scoring points are clustered into discrete pockets, ranked by aggregate score

Understanding druggability scores

Each predicted pocket receives a druggability score between 0 and 1:

Score rangeInterpretationWhat it means
0.80 – 1.00Highly druggableDeep, well-defined cavity. Strong candidate for small-molecule binding.
0.50 – 0.79Moderately druggableReasonable pocket but may be shallow or partially exposed.
0.20 – 0.49ChallengingFlat or exposed surface. May require fragment-based approaches.
0.00 – 0.19UnlikelyNot a viable drug binding site with current methods.

How to interpret pocket residues

Each pocket is defined by the amino acid residues that form its walls. OpenDDE shows you the residue composition:

  • Hydrophobic residues (Leu, Ile, Val, Phe) — form the “greasy” interior of the pocket. More hydrophobic = better for small-molecule binding.
  • Polar residues (Ser, Thr, Asn, Gln) — provide hydrogen bonding partners for drug design.
  • Charged residues (Asp, Glu, Lys, Arg) — can form salt bridges with charged drug groups.
  • Aromatic residues (Phe, Tyr, Trp) — enable pi-stacking interactions with aromatic drug rings.

Limitations and caveats

  • Static structures — P2Rank operates on a single 3D snapshot. Proteins are dynamic; some pockets only open during conformational changes (cryptic sites).
  • Allosteric sites — P2Rank focuses on orthosteric (active site) pockets. Allosteric sites may be ranked lower even if therapeutically relevant.
  • Predicted structures — When using AlphaFold-predicted structures (vs. experimental crystal structures), pocket predictions may be less reliable in low-confidence regions.
  • Protein-protein interfaces — Pockets at protein-protein interaction sites may not be detected as traditional small-molecule binding sites.

Next: Ligand intelligence →