Constructing an open database of barrier heights for diverse elementary organic reactions

Understanding chemical mechanism is a cornerstone of reaction design, development, and optimization. Models which can predict reaction mechanism have the potential to elucidate plausible reaction pathways, detect problematic steps, identify side products and byproducts, and propose unprecedented synthetic methods. Chemical reaction networks, or CRNs, are often used to depict possible mechanistic pathways. CRNs can be constructed on the fly using enhanced sampling techniques, but the process is limited by the computational time required to find each elementary reaction. We can sidestep these computational demands by building CRNs using a sufficiently comprehensive list of possible elementary steps. Herein, we discuss our efforts toward the construction of a dataset containing diverse elementary organic reactions. To further predict which of multiple possible elementary steps will occur, we outline a process for calculating solution-state reaction barriers using quantum chemical methods. The first iteration of this work will be confined to main group chemistry. Finally, we describe progress toward constructing CRNs using machine learning models trained on this barrier height data. Our strategies focus on developing a process which is robust to different types of elementary reactions, thereby enabling growth of an open dataset for elementary step reactivity.