How accurate is the description of ligand–protein interactions by a hybrid QM/MM approach?

During the last decades, the application of hybrid quantum mechanical/molecular mechanical (QM/MM) methods has been extended to the field of drug design. In principle, the approximate QM/MM approach offers a more complete description of drug–receptor non-covalent interactions. This is especially true when charge or proton transfer, chelation of metal ions or strong polarization of ligand and protein or surface chemical groups are involved. The aim of this work was to assess the accuracy of calculated non-covalent ligand–protein interaction energies ( ∆ E int QM / MM ) obtained by the hybrid QM/MM approach employed in QSite/Jaguar of Schrödinger’s Small-Molecule Drug Discovery Suite on a set of small-molecule model systems when compared to rigorous QM calculations. The QM/MM approach was used at the density functional theory (DFT) level of theory with 6-31G* basis set, hybrid B3LYP functional and OPLS-2005 force field (DFT-B3LYP/6-31G*//OPLS-2005), a popular combination frequently used in studies on larger and complex biological systems such as drug–receptor complexes. In this work, we did not attempt to compute the most precise interaction energies of the model systems. We rather tried to assess the performance of the approximate QM/MM vs. full QM approach at the same computationally accessible level. For effective use of the QM/MM approach it is essential to select an appropriate QM region of the studied systems. To aid the selection of specific protein residues or functional groups to be included in the QM region, we evaluated the effect of its size, composition and symmetry on the accuracy of the QM/MM calculated ∆ E int QM / MM . This was performed by means of a set of model clusters with well-defined configurations, which mimic the basic types of non-covalent interactions in proteins. Based on these systematic quantitative comparisons, recommendations for the addition of chemical groups or protein residues into the QM region are proposed for the popular DFT-B3LYP/6-31G*//OPLS-2005 QM/MM approach, leading to a more realistic description of ligand–protein interactions. These guidelines can have a significant bearing on computational drug or material research employing hybrid QM/MM methods by providing an estimate of the accuracy that can be expected from QM/MM studies. Graphical abstract An approximate hybrid QM/MM approach at the DFT-B3LYP/6-31G*//OPLS-2005 level was systematically assessed for the accuracy of description of non-covalent interactions o