Reactants, products, and transition states of elementary chemical reactions based on quantum chemistry

Abstract

Reaction times, activation energies, branching ratios, yields, and many other quantitative attributes are important for precise organic syntheses and generating detailed reaction mechanisms. Often, it would be useful to be able to classify proposed reactions as fast or slow. However, quantitative chemical reaction data, especially for atom-mapped reactions, are difficult to find in existing databases. Therefore, we used automated potential energy surface exploration to generate 12,000 organic reactions involving H, C, N, and O atoms calculated at the ωB97X-D3/def2-TZVP quantum chemistry level. We report the results of geometry optimizations and frequency calculations for reactants, products, and transition states of all reactions. Additionally, we extracted atom-mapped reaction SMILES, activation energies, and enthalpies of reaction. We believe that this data will accelerate progress in automated methods for organic synthesis and reaction mechanism generation—for example, by enabling the development of novel machine learning models for quantitative reaction prediction. Background & Summary Rapid advancements in computational methods for chemical synthesis planning and automated reaction mechanism generation, especially in the area of machine learning, are causing a significant shift in how such problems are tackled. Deep learning approaches are replacing conventional quantitative structure-activity relationships often based on support vector machines, decision trees, or linear methods like partial least squares1,2. These new systems are becoming widely available for computer-aided retrosynthesis3, reaction outcome prediction3, high-throughput virtual screening4, and more general molecular property prediction5,6. Computational approaches are also increasingly common in reaction mechanism generation due to the large number of species and reactions that are generally required for accurate descriptions of phenomena like pyrolysis, combustion, and atmospheric oxidation7–9. Frequently, this involves characterizing chemical pathways with quantum chemistry8, but deep learning methods have also recently been applied to estimate thermochemistry during mechanism generation10,11. While computers already outperform humans at qualitatively predicting reaction products12,13 and successful yield predictions have been demonstrated for limited datasets14,15, quantitative reaction information is still elusive in large databases like Reaxys16, Pistachio17, and the United States Patent and Trademark Office database18. Reaction yield, time, and some quantitative conditions like temperature are sometimes available, but there is usually no information on reaction kinetics. If such data were available, calculation of derived properties— such as minimum reaction times and branching ratios—would be possible. Our goal is to provide a quantitative dataset of reactions that enables the calculation of such data and can lead to more efficient drug design and help in deciding which reactions are important in mechanism generation. Computationally generating a dataset of reactions is significantly more complex than only calculating stable equilibrium structures because transition states (TSs) of chemical reactions cannot be enumerated in the same manner as stable molecules. Even if the reactant and product structures are known, the exact TS geometry has to be found via a human-guided search or with expensive automated TS finding methods. Here, we use automated potential energy surface exploration to generate the dataset of reactions, which has been shown to be successful in cases when many reaction pathways have to be evaluated19–21. More specifically, we rely on the growing string method22 to automatically optimize reaction paths and TSs. We report quantum chemical data on more than 16,000 reactions in the form of reactants, products, and TSs at the B97-D3/def2-mSVP level of theory, and 12,000 reactions at the ωB97X-D3/def2-TZVP level of theory. The data include the raw output from geometry optimizations and frequency calculations in addition to atom-mapped SMILES, activation energies, and enthalpies of reaction. All reactions are gas-phase calculations involving up to seven carbon, oxygen, or nitrogen atoms per molecule. The reactants are sampled from GDB-7, a subset of GDB-1724, meaning that all reactions have a unimolecular reactant but potentially multi-molecular products. Figure 1 illustrates the dataset generation process and the resulting space of reactions in terms of their activation energies and enthalpies of reaction. 0 50 100 150 200 Ea (kcal/mol) 150 100 50 0 50 100 150