Validation of a computational docking methodology to identify the non-covalent binding site of ligands to DNA

Despite the biomedical consequences of carcinogen-DNA interactions and the potential of DNA as a drug target in medicinal chemistry, only a small number of studies have validated or used docking methods for the prediction of the physical binding of small molecules to DNA. Knowledge of the DNA-physically-bound ligand geometry can lead to the elucidation of the molecular-level mechanism of drugs as well as predicting the subsequent chemical interactions that lead to DNA damage from carcinogens. We sought to validate AutoDock 4.2, a docking method that includes a physics-based free energy function and a Lamarckian Genetic Algorithm, for the prediction of ligand geometries upon physical binding to DNA. We performed simulations by systematically changing the length of the search process for a comprehensive set of 32 ligand-DNA molecular systems with different physico-chemical properties, and we used a free-energy-based convergence criterion to terminate our simulations. For 11 out of 28 molecular systems for which convergence was achieved, the lowest binding free energy geometries were within 2 Å of the experimentally determined geometry. Considering all predicted sites with free energy changes within 20% of the lowest binding free energy site, we found a site within 2 Å of the experimentally determined geometry for 24 out of the 28 systems. However, the predicted hydrogen bonding interactions were different for most molecular systems compared to the same interactions in the experimentally determined geometry. We discuss reasons for the successes and failures, implications, and the importance of ensuring an adequate search in docking calculations. Overall, we concluded that AutoDock 4.2 can be used to predict the non-covalent binding geometry of a small molecule to DNA with some limitations.