(346ah) Towards Predicting Kinetics in Systems of Coupled Reactions with Deep Neural Networks

The computational cost of modeling kinetics atomistically grows with the system’s size.^1,2 Therefore, this approach becomes unfeasible for large networks of reactions and few efforts have been made in this direction.³ Recently there has been interest in using machine learning models to predict kinetics.⁴ Indeed, trained deep neural networks (DNNs) have the potential to alleviate the computational cost of atomistic simulations⁵ by making predictions based on easily computable static properties such as optimized reactant and product geometries and ground state energies.⁶ In this context, we aim to understand whether machine-learned rate constants can accurately describe the populations of chemical species in a network of reactions. Hence, we trained DNN models using existing databases of gas-phase kinetic data to predict reaction rate constants. Optimal DNN hyperparameters were identified by exploring the structure of the model, and input features were selected according to their importance in predicting the rate. We then looked at a small set of coupled gas-phase reactions as a testing set and used our trained DNN to predict the rate constants. The accuracy of the DNN at the level of network populations was evaluated by comparing the exact model with the one predicted by the DNN.

1. Ju, L. P., Han, K. L. & Zhang, J. Z. H. Global dynamics and transition state theories: Comparative study of reaction rate constants for gas-phase chemical reactions. Journal of Computational Chemistry (2009) .

2. Sumathi, R. & Green, W. H. A priori rate constants for kinetic modeling. Theoretical Chemistry Accounts (2002) .

3. Wang, L. P. et al. Discovering chemistry with an ab initio nanoreactor. Nat. Chem. (2014) .

4. Gao, H. et al. Using Machine Learning to Predict Suitable Conditions for Organic Reactions. ACS Cent. Sci. 4, 1465–1476 (2018).

5. Valleau, S. & Martínez, T. J. Reaction Dynamics of Cyanohydrins with Hydrosulfide in Water. J. Phys. Chem. A (2019) .

6. Collins, C. R., Gordon, G. J., Von Lilienfeld, O. A. & Yaron, D. J. Constant size descriptors for accurate machine learning models of molecular properties. J. Chem. Phys. 148, (2018).