An automated intermolecular reaction discovery approach relying on heuristic atom-partitioned frontier orbital features

Large-scale computational exploration of reactions offers a new perspective for un- derstanding chemical reaction processes. However, it often relies on chemical intuition and extensive manual effort. We introduce here a novel algorithm for fast exploration of possible products in bimolecular reactions. The algorithm is based on atomistic features derived from the inexpensive electronic structure theory calculations. With the electronic structure as the sole required input from reactants, the reaction search includes the assessment of reactivity sites, the construction of corresponding trial reaction coordinates, and their exploration, all of which can be carried out autonomously. We demonstrate that our algorithm significantly reduces the manual effort involved in chemical reaction exploration using as reference a large set of prototype reactions obtained from textbooks and published databases. Furthermore, a new product exploration approach based on a harmonic force field optimization achieves around 97% success in reproducing reference reaction outcomes, while keeping the computational cost low, thus, ensuring both reliability and efficiency. The entire reaction exploration workflow has been implemented in a self-developed reaction exploration package called aRST.