Quasi-atomistic receptor modeling. A bridge between 3D QSAR and receptor fitting.

Quasi-atomistic receptor modeling bridges 3D QSAR and receptor modeling by combining receptor surface models populated with atomistic properties (hydrogen bonds, salt bridges, aromatic and aliphatic regions, solvent) with individually fitted receptor envelopes--simulating a flexible receptor cavity, adapted by means of an induced fit. To mimic amino-acid residues capable to engage in differently directed H-bonds with different ligand molecules at the true biological receptor, our approach includes H-bond flip-flop particles which, depending on a given ligand molecule, can simultaneously act as a H-bond donor and a H-bond acceptor. The use of a directional force field for hydrogen bonds allows for the simulation of ligand selectivity, including the discrimination of stereoisomers. Based on a series of ligand molecules with individually fitted receptor envelopes, the software Quasar allows to generate a family of receptor models by means of genetic algorithms combined with cross-validation protocols. Our concept has been used to derive semi-quantitative structure--activity relationships for a series of six receptor systems, including the beta 2-adrenergic, dopaminergic, aryl hydrocarbon, cannabinoid, neurokinin-1, and HIV protease, respectively. For these systems, quasi-atomistic receptor modeling, was able to predict the relative free energies of ligand binding of an independent set of test ligands within 0.55 to 0.94 kcal/mol of their experimental value, corresponding to an uncertainty in the binding affinity of a factor of 2.5 to 5.0.