Ring-opening reactions are the key factors determining the major products from various chemical reactions such as ring-opening polymerization and natural product synthesis. Combustion properties of cyclic fuel compounds are also governed by ring-opening reactions occurring at the initiation stage of fuel ignition. The determination of ring-opening bond dissociation enthalpies (BDEs) informs reactivity and site-selectivity of ring-opening reactions. However, ring-opening reactions form diradical species, and their accurate calculations usually require high-level quantum-mechanical methods (e.g., composite and multireference methods) which are time-consuming, thus demanding the development of an efficient machine learning (ML) prediction method. Here, we will introduce our machine learning model to predict bond dissociation enthalpies of ring-opening reactions with its application to various fields of chemistry. Our approach for the development is as follows; the first step is to collect experimental thermochemistry data pertaining to ring-opening BDEs, such as standard enthalpy of formation (ΔHf) of the cyclic compounds. Since experimental ΔHf values of diradicals are very rare, their ΔHf’s are derived from a set of isodesmic reactions. Next, we assess the accuracy of various theoretical methods including density functionals and multireference methods, by calculating ΔHf’s of diradicals and comparing them with the experimentally derived ones. Based on the assessment results, we select the most efficient density functional which shows the best compromise between accuracy and computational cost. The BDE database is then built by calculations using the chosen method. The molecules in the database will include those in our published BDE database (BDE-db, 16,868 cyclic compounds, and 74,694 bonds in a ring) and other compounds of interest. As a final step, the ML model will be developed by training BDEs in the established database. We will show applications of our model such as (i) controlling site-selectivity by tuning ring strain and bond strength with different functional groups, and (ii) identification of the weakest bond in a ring and its correlation with fuel characteristics.

Development of a machine learning model to predict bond dissociation enthalpies of ring-opening reactions

ACS Fall 2021

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