Binding free energy estimation for protein-ligand complex based on MM-PBSA with various partial charge models

An accurate estimation of binding free energy between protein and ligand, is one of the most important issues in the drug discovery process. However, it is an arduous and hard process to obtain accurate energy, especially the experimentally relevant free energies for protein-ligand in solution, including a proper treatment of the long-range electrostatics and solvation effects that are involved in optimization of atomic net charges and so on. In this study, the impacts of the various atomic net charge models were considered, and their effects on binding free energy profiles also were investigated. The methods were tested on: the 30 structurally diverse ligands of diverse protein complexes, the 14 structurally diverse ligands of the protein kinase B (PKB) and the 10 structurally diverse ligands of the cyclin-dependent kinases 2 (CDK2) with measured affinities. The tested charges were calculated based on AM1 (Austin Method, version 1) - BCC (bond charge correction), MNDO (modified neglect of diatomic differential overlay), PM5 (Parameterisation Model, version 5), MUL (Mulliken), CM2 (Charge Model 2), CM3 (Charge Model 3), RESP (restrained electrostatic potential) and QM/MM (quantum mechanics/molecular mechanics) models. Our findings showed that the MNDO charge model was best propitious for PKB system and QM/MM for CDK2, whereas none of any given models was suited for the diverse ligands of diverse protein complexes. The trends of MM-PBSA binding free energies using all charge models were in good accord with experimental results for CDK2 but not for PKB in most cases. Considering the above results, particular attention should be paid to the ligand-charge and maybe protein-charge during the estimation of accurate binding free energies in drug design.