The homolytic cleavage of covalent bonds is of great importance to many chemistry areas, especially in metabolism, combustion and catalysis. An instant predictor for bond dissociation energies (BDEs) of a large number of molecules that doesn’t require specific devices (super-cluster) is therefore of great value to a broad cross-section of chemists. However, due to the absence of databases for BDEs, it is quite difficult to develop and evaluate such tools. Herein, we summarize our progress in building a BDE database, which significantly facilitates accessing, querying and analysis for a large amount of BDE data. This BDE database is a SQL database and is built on the Atomic Simulation Environment Data Base. A workflow was built to automatically grab molecules from PubChem dataset, optimize molecules/radicals, calculate BDEs, and populate the BDE database. Furthermore, the constructed database is used as training set for several different machine learning models to predict BDEs. The best model, modified Schnet (a continuous-filter convolutional neural network), provides comparable accuracy to DFT calculations at significantly lower costs. Using the selected model, a web-based instant BDE predictor will be built.

BDEDB: A bond-dissociation energy database and instant prediction

2019 Spring National Meeting

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