AROS: Alchemical Rosetta Enables Accurate and Efficient Relative Binding Free Energy Calculations

Abstract

Computer aided drug design has accelerated drug-to-market timelines through advancements in virtual screening tools and in silico lead optimization workflows. Rosetta’s GALigandDock and RosettaGenFF forcefield have been characterized together as a robust virtual screening platform but remain untested against specific lead optimization tasks. Additionally, Rosetta currently lacks framework to execute alchemical perturbation simulations which are used by modern best-in-class lead optimization workflows. In this dissertation, I describe my efforts to position Rosetta as a platform for efficient in silico lead optimization in two ways. First, I formally benchmark GALigandDock using RosettaGenFF on lead optimization tasks and highlight its ability as a rapid evaluation tool. Second, I develop and benchmark Alchemical Rosetta, a new framework that required methodological changes to allow alchemical perturbation simulations within Rosetta. I do so by augmenting Rosetta to correctly weigh partial ligand occupancy in the binding pocket and move shared substructure in lockstep during the simulation. Alchemical Rosetta offers faster convergence and shorter simulation times in combination with a greatly reduced computational cost compared to current state of the art methods.