ModBind _dG : A simulation-based absolute predictor of free energy of binding based on population reweighting

Abstract

Recently, we described ModBind, a powerful simulation-based predictor of ligand off-rates and binding free energy. Here, we describe an update to the method—ModBind dG —that allows for the prediction of absolute free energies of binding. ModBind dG is a two-state model for predicting free energy of binding through population-based reweighting of accelerated sampling. We demonstrate theoretically and in practice that ModBind dG can accurately predict the absolute binding free energies of small molecules binding to relevant protein targets. Additionally, we show performance enhancements to the method where ModBind dG can predict up to 2,000× more compounds per day compared to state-of-the-art free energy methods. We show that the method can make accurate predictions on validation datasets as well as active drug discovery programs in our own pipeline. Furthermore, we describe the utility of ModBind dG in a prospective virtual screen that enabled the discovery of multiple chemotypes for a previously “undruggable” protein target. ModBind dG represents an opportunity to significantly impact computational drug discovery, making rigorous physics-based screening of hundreds of thousands to millions of compounds accessible to the entire pharmaceutical community.