Machine learning potentials have great promise in the design of fast, accurate force fields for a wide range of chemical applications including the study of mechanisms of enzyme reactions to guide enzyme design, and the prediction of ligand-protein binding affinities for drug discovery.  In this talk new range-corrected deep-learning potentials (DPRc) are discussed as a robust method for enhancing the accuracy and performance of free energy simulations.  Specifically, we develop models to study complex mechanisms of RNA enzymes with accuracy comparable to ab initio quantum mechanical/molecular mechanical potentials (with over 400X speed-up), and use path integral molecular dynamics (PIMD) to predict and interpret experimentally measured kinetic isotope effects (KIEs).  Simulations are further enhanced by a new generalized weighted thermodynamic perturbation (gwTP) methods for estimating the free energy surface of an expensive ``high-level'' potential energy function from simulations performed with inexpensive ``low-level'' reference potentials.  We further describe the development of a new Quantum Deep Potential Interaction Energy (QDπ) model for use in alchemical free energy simulations to enhance prediction of ligand-protein binding affinities.  These methods are made more affordable in the high-sampling regime through use of reference potential “book-ending” free energy methods.

Range-corrected deep-learning potentials for free energy simulations applied to enzyme design and drug discovery

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.