Identification of binding pockets in protein structures using a knowledge-based potential derived from local structural similarities

We have developed PDBinder, a novel method for the prediction of ligand binding sites in protein structures. Our approach is based on the observation that unrelated binding sites share small structural motifs that bind the same chemical fragments irrespective of the nature of the ligand molecule as whole. PDBinder compares a query protein against a library of binding and non-binding protein surface regions derived from the Protein Data Bank (PDB). The results of the comparison are used to derive a propensity value for each residue which is correlated with the likelihood that the residue is part of a ligand binding site.

PDBinder has been trained on a non-redundant set of 1356 high-quality protein-ligand complexes and tested on a set of 267 holo and apo complex pairs. We obtained an MCC of 0.327 on the holo set with a PPV of 0.436 while on the apo set we achieved an MCC of 0.289 and a PPV of 0.402. The good performance of PDBinder on the unbound proteins is extremely important for real-world applications where the location of the binding site is unknown.
We show that the performance of PDBinder is superior to that of other methods both in the prediction of specific binding residues and in the identification of which cavity in the structure is most likely to host a ligand binding site. Moreover, since our innovative approach is orthogonal to those used in existing methods, the propensity value assigned by PDBinder can be integrated in other algorithms, further increasing the final performance.

No relevant conflicts of interest declared.