PDBinder

PDBinder predicts small ligand binding sites in protein structures by comparing query proteins to a library of protein surface regions from the Protein Data Bank (PDB) and computing residue propensity values that indicate likelihood of ligand interaction.

Key Features:

• Knowledge-based library: Uses a library of protein surface regions derived from the Protein Data Bank (PDB) that distinguishes binding and non-binding areas for comparative analysis.
• Residue propensity calculation: Derives a propensity value for each residue that correlates with the likelihood of participation in a ligand binding site.
• Dual prediction outputs: Predicts specific ligand-binding residues and identifies which surface clefts harbor binding sites.
• Training dataset: Trained on a non-redundant set of 1,356 high-quality protein–ligand complexes.
• Validation metrics: Tested on 239 holo/apo complex pairs with reported Matthews Correlation Coefficient (MCC) of 0.313 and Positive Predictive Value (PPV) of 0.413 on the holo set, and MCC of 0.271 with PPV of 0.372 on the apo set.
• Integration compatibility: Produces propensity values that are orthogonal to other methods and can be integrated into additional algorithms.

Scientific Applications:

• Structural annotation: Annotation of protein function by locating small-molecule binding sites on proteins with known structures.
• Drug discovery and design: Identification of potential small-molecule binding sites in both bound (holo) and unbound (apo) protein structures to inform ligand design.
• Algorithm development and model refinement: Provision of propensity values for incorporation into other predictive methods to improve binding-site prediction.

Methodology:

Compares a query protein to a PDB-derived library of binding and non-binding surface regions and computes per-residue propensity values; method was trained on 1,356 non-redundant protein–ligand complexes and tested on 239 holo/apo pairs with reported MCC and PPV values.