To model reactions in solution, we have developed a neural network method for QM/MM calculation. Using neural network, the potential energy of any configuration along the reaction path can be predicted at the ab initio QM/MM level based on the semi-empirical QM/MM simulations. The free-energy profile obtained from the semi-empirical QM/MM simulation is corrected to the ab intio QM/MM level by using the potential energies predicted by the constructed neural network. The results are in excellent accordance with the reference data. Our method shows a speed-up of one or two orders of magnitude, suggesting that the neural network method combined with the semi-empirical QM/MM calculation is an efficient and reliable strategy for biochemical simulations. We also developed a dynamic scheme of QM/MM-NN for direct MD simulations, and a novel neural network-based force field for water based on training with high-level ab initio theory. The water force field was built based on an electrostatically embedded many-body expansion method truncated at binary interactions. Neural networks were trained to represent electrostatically embedded one-body and two-body interactions, which require as input only one and two water molecule calculations at a high level of ab initio electronic structure method.  This force field approach is general and has been extended to polypeptides.  Our results demonstrate that the electrostatically embedded many-body expansion combined with neural network provides a promising and systematic way to build next-generation force fields at high accuracy and low computational costs, especially for large systems.<br/>

Simulations and force fields with quantum mechanics/molecular mechanics and machine learning

2019 Fall National Meeting

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