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

Area of Science:

• Computational chemistry
• Biophysics
• Enzyme kinetics

Background:

• Accurate modeling of enzyme reactions using quantum mechanics/molecular mechanics (QM/MM) simulations is hindered by challenges in describing electrostatic coupling.
• Existing methods struggle to efficiently and accurately capture the interactions between QM and MM subsystems.

Purpose of the Study:

• To develop a novel machine learning-based method for improved QM/MM simulations of enzyme reactions.
• To accurately model electrostatic coupling and predict enzyme activity and mutant properties.

Main Methods:

• Proposed a reweighting mechanic embedding (ME) recursively embedded atom neural network (REANN) method.
• Trained potential energy and point charges of the QM subsystem in vacuo using charge equilibration.
• Corrected MM polarization effects using thermodynamic perturbation after molecular dynamics simulations.

Main Results:

• Successfully reproduced free energy curves for aspirin acylation of COX-1 and COX-2 with chemical accuracy.
• Accurately predicted the free energy barrier for a COX-2 mutant (R513A) with <0.5 kcal mol⁻¹ difference.
• Achieved an 80-fold speedup in calculations compared to traditional QM/MM methods.

Conclusions:

• The REANN method provides accurate and rapid predictions of enzyme activity and mutant effects.
• This approach significantly enhances the efficiency of QM/MM simulations for biochemical reactions.
• The method holds promise for future applications in virtual screening and drug discovery.