Cinematic Simulation of Substrate-to-Product Chemical Reactions with Chemical Accuracy Using QuantaMind MD

Abstract

Polyethylene terephthalate hydrolases (PETases) are enzymes that catalyze the breakdown of PET plastic. Previous studies have employed classical molecular dynamics (MD) and quantum mechanical/molecular mechanical (QM/MM) simulations to probe the catalytic mechanisms of the PETase-catalyzed reaction. In this work, we apply QuantaMind, a deep learning-based machine learning force field (MLFF) trained on diverse chemical systems and transition-state energies at density functional theory (DFT)-level, to simulate the complete catalytic cycle of the PETase reaction. Our simulations capture key proton transfer events and the stabilization role of the oxyanion hole in PETase, in agreement with previous computational studies, without introducing artificial biases to the interactions. The simulations also provide estimates of free energy barriers for each reaction step with uncertainty estimation, consistent with experimental values and previous QM/MM studies. To our knowledge, this work represents the first complete DFT-level MD simulation of a biomolecular enzymatic reaction using DL-based MLFFs.