Abstract
Ligand-protein docking aims to predict how a ligand binds to a biological macromolecule and is a fundamental technique in structure-based computer-aided drug design. However, accurately modeling covalent binding, metal coordination, and polarization effects remains challenging for classical docking algorithms. Here, we present an extension of our Attracting Cavities docking algorithm that enables hybrid quantum mechanical/molecular mechanical (QM/MM) calculations at various levels of theory. To evaluate its performance, we benchmarked the method on three diverse datasets covering non-covalent drug-target complexes, covalent complexes, and hemoprotein complexes, using both semi-empirical and density functional theory approaches. The results demonstrate that QM/MM docking is especially advantageous for metal-binding complexes, where the fast semi-empirical PM7 method yields a significant improvement over classical docking. When describing the active site residues at the density functional theory level, dispersion corrections are crucial for meaningful energies. Overall, the QM/MM method outperforms the classical approach for metalloproteins, performs comparably for covalent complexes, and shows slightly lower success rates for non-covalent complexes.