Prediction of Drug-Target Unbinding Kinetics by Molecular Simulations

Drug design and discovery is a time-consuming and resource-intensive process. Several factors contribute to the high probability of failure during drug design process, among which a still poor comprehension of drug-target interactions. Binding affinity has historically been assumed as valid indicator to evaluate a drug candidate. However, recent studies demonstrated the relevance of other drug-receptor interaction properties in order to maximise in vivo pharmacological activity. In the last years, greater consideration was given to the drug residence time as better efficacy predictor than the binding affinity. The drug residence time concept involves a deeper understanding of binding and unbinding mechanisms of ligand-protein complexes, including the characterisation of kinetic rate constants. Despite several experimental techniques are available to determine (un)binding kinetic constants, efficient computational approaches are missing. This thesis project proposes a novel computational method able to provide a reliable protein-ligand unbinding kinetics prediction. A protocol of enhanced sampling molecular dynamics simulations for unbinding energy estimation has been combined with a procedure of dissociation pathways prediction. Indeed, the rupture of drug-target stable interactions has been carried out by following well-defined escape pathways. The energy required to remove the ligand from the binding site has been estimated and used to evaluate a possible correlation with the experimentally determined unbinding times. The methodology has been tested on pharmacologically relevant biological systems consisting in the molecular chaperone HSP90, an important cancer target, complexed with some of its main inhibitors. The present approach aids the investigation of ligand-protein dissociation processes, while being easily implementable. Further studies are needed to extend the scope of the protocol and overcome limitations related to the unbinding pathways’ prediction. Importantly, compared to the state of the art of scientific literature, this method offers a less computationally demanding alternative.