UCLA

Chapter 1 describes the development of Auto-QChem, an automated, high-throughput and end-to-end density functional theory (DFT) calculation tool that can generate quantum chemical descriptors for organic molecules. We discuss in detail the design and implementation of Auto-QChem, as well as its current functionalities. We also review literature examples in synthetic organic chemistry where Auto-QChem-derived descriptors were applied in machine learning (ML) models to accelerate methodology development. Chapter 2 describes the design, implementation and application of reinforcement learning bandit optimization models in chemistry reaction optimization, where generally applicable reaction conditions were identified via efficient condition sampling and evaluation of experimental feedback. In addition to performance benchmarking on existing reaction datasets in literature, we also experimentally investigated a palladium-catalyzed imidazole C–H arylation reaction, an aniline amide coupling reaction and a phenol alkylation reaction. In all three cases, bandit optimization models identified most generally applicable yet not well studied reaction conditions for the respective reaction. Chapter 3 describes the discovery and characterization of multiple N-(hetero)aryl, N-benzyl and N-alkyl derivatives of the 9-mesityl-3,6-di-tert-butyl-10-phenyl acridinium photocatalyst. The catalytic performances of these catalysts as photo-oxidant or photo-reductant (via in situ generated acridine radical) were compared in three model reactions. We also identified improved catalytic conditions for a previously reported cyanoarene-catalyzed nucleophilic amination reaction using a synthesized N-cycloheptyl acridinium catalyst with up to 98% reaction yield.