Modeling Enzyme Reaction and Mutation by Direct Machine Learning/Molecular Mechanics Simulations

Accurately modeling enzyme reactions through direct machine learning/molecular mechanics simulations remains challenging in describing the electrostatic coupling between the QM and MM subsystems. In this work, we proposed a reweighting ME (mechanic embedding) REANN (recursively embedded atom neural network) method that trains the potential and point charges of the QM subsystem in vacuo. The charge equilibration approach has been encoded into REANN to ensure conservation of the total charge of the QM subsystem. Electrostatic coupling is measured by point charges, and the polarization of the MM subsystem on the coupling can be corrected by thermodynamic perturbation after molecular dynamics simulations. We first constructed the REANN surfaces of potential energy and charges for the acylation of cyclooxygenase-1 (COX-1) and cyclooxygenase-2 (COX-2) by aspirin. These surfaces allowed us to reproduce the free energy curves of B3LYP/MM-MD with a chemical accuracy. Subsequently, they were successfully applied to R513A of COX-2, reproducing the free energy barrier simulated by B3LYP/MM MD with a difference of less than 0.5 kcal mol^–1 and a speedup of 80-fold, revealing our method can predict the activity of mutants accurately and rapidly. This method is expected to be applied in virtual screening in the future.