Comparative binding energy analysis for binding affinity and target selectivity prediction

A major challenge in drug design is to obtain compounds that bind selectively to their target receptors and do not cause side-effects by binding to other similar receptors. Here, we investigate strategies for applying COMBINE (COMparative BINding Energy) analysis, in conjunction with PIPSA (Protein Interaction Property Similarity Analysis) and ligand docking methods, to address this problem. We evaluate these approaches by application to diverse sets of inhibitors of three structurally related serine proteases of medical relevance: thrombin, trypsin, and urokinase-type plasminogen activator (uPA). We generated target-specific scoring functions (COMBINE models) for the three targets using training sets of ligands with known inhibition constants and structures of their receptor-ligand complexes. These COMBINE models were compared with the PIPSA results and experimental data on receptor selectivity. These scoring functions highlight the ligand-receptor interactions that are particularly important for binding specificity for the different targets. To predict target selectivity in virtual screening, compounds were docked into the three protein binding sites using the program GOLD and the docking solutions were re-ranked with the target-specific scoring functions and computed electrostatic binding free energies. Limits in the accuracy of some of the docking solutions and difficulties in scoring them adversely affected the predictive ability of the target specific scoring functions. Nevertheless, the target-specific scoring functions enabled the selectivity of ligands to thrombin versus trypsin and uPA to be predicted.