Computational prediction of binding affinity for CYP1A2-ligand complexes using empirical free energy calculations.

Predicting binding affinities for receptor-ligand complexes is still one of the challenging processes in computational structure-based ligand design. Many computational methods have been developed to achieve this goal, such as docking and scoring methods, the linear interaction energy (LIE) method, and methods based on statistical mechanics. In the present investigation, we started from an LIE model to predict the binding free energy of structurally diverse compounds of cytochrome P450 1A2 ligands, one of the important human metabolizing isoforms of the cytochrome P450 family. The data set includes both substrates and inhibitors. It appears that the electrostatic contribution to the binding free energy becomes negligible in this particular protein and a simple empirical model was derived, based on a training set of eight compounds. The root mean square error for the training set was 3.7 kJ/mol. Subsequent application of the model to an external test set gives an error of 2.1 kJ/mol, which is remarkably good, considering the simplicity of the model. The structures of the protein-ligand interactions are further analyzed, again demonstrating the large versatility and plasticity of the cytochrome P450 active site.