Abstract

The majority of biological processes involve the association of proteins or binding of other ligands to proteins. The accurate prediction of putative binding sites on the protein surface can be very helpful for rational drug design on target proteins of medical relevance, for predicting the geometry of protein-protein as well as protein-ligand complexes and for evaluating the tendency of proteins to aggregate or oligomerize. A variety of computational methods to rapidly predict protein-protein binding interfaces or binding sites for small drug-like molecules have been developed in recent years. The principles of methods available for protein interface and pocket detection are summarized, including approaches based on sequence conservation, as well as geometric and physicochemical surface properties. The performance of several Web-accessible methods for ligand binding site prediction has been compared using protein structures in bound and unbound conformation and homology modeled proteins. All methods tested gave very promising predictions even on unbound and homology modeled protein structures, thus indicating that current methods are robust in relation to modest conformational changes associated with the ligand binding process.