NCASD/UCI Chemistry Dept. postdoc, Gianmarc Grazioli, our collaborator from the UCI Chemistry Dept., Saswata Roy, and PI, Carter T. Butts have published a machine learning-based methodology for the prediction of chemical reaction products from atomistic computer simulations entitled “Predicting Reaction Products and Automating Reactive Trajectory Characterization in Molecular Simulations with Support Vector Machines,” at the Journal of Chemical Information and Modeling (JCIM). In contrast to the more common “black box” type machine learning methodologies for analyzing chemical simulation data, this SVM-based methodology allows for mechanistic insight to be gleaned from further analysis of the positioning of the phase space points used to train the SVM with respect to the separatrices. In addition to inferring mechanistic details about multiple-pathway chemical reactions, our method can also be used to increase reactive trajectory sampling efficiency in molecular simulations via rejection sampling, with trajectories leading to undesired product states being identified and terminated early in the simulation rather than being carried to completion. In the article, we demonstrate our methodology using low dimensional heuristic models and then apply it to ab-initio computer simulations of the photodissociation of acetaldehyde, an important chemical system in atmospheric chemistry.
link to the article:https://pubs.acs.org/doi/10.1021/acs.jcim.9b00134