Deciphering common failures in molecular docking of ligand-protein complexes

Common failures in predicting crystal structures of ligand-protein complexe s are investigated for three ligand-protein systems by a combined thermodyn amic and kinetic analysis of the binding energy landscapes. Misdocked predi ctions in ligand-protein docking are classified as 'soft' and 'hard' failur es. While a soft failure arises when the search algorithm is unable to find the global energy minimum corresponding to the crystal structure, a hard f ailure results from a flaw of the energy function to qualify the crystal st ructure as the predicted lowest energy conformation in docking simulations. We find that neither the determination of a single structure with the lowe st energy nor finding the most common binding mode is sufficient to predict crystal structures of the complexes, which belong to the category of hard failures. In a proposed hierarchical approach, structural similarity cluste ring of the conformations, generated from equilibrium simulations with the simplified energy function, is followed by energy refinement with the AMBER force field. This protocol, that involves a hierarchy of energy functions, resolves some common failures in ligand-protein docking and detects crysta llographic binding modes that were not found during docking simulations.