The discovery and optimization of homogeneous catalysts remain bottlenecks in the development of efficient and selective chemical transformations. Here we report Flex-Cat, a closed-loop autonomous catalysis platform that couples parallel miniaturized batch reactors for pressurized gas–liquid chemistry with a hierarchical, plate-constrained Bayesian optimization framework for mixed discrete (ligand identity) and continuous (process) variables. Using rhodium-catalyzed hydroformylation of propylene and a chemically diverse library of phosphorus-based ligands, Flex-Cat conducts 680 experiments across three multi-objective optimization campaigns targeting branched, linear, and tunable aldehyde regioselectivity. Across three campaigns targeting branched, linear, and regioselectivity flexibility, Flex-Cat identified ligand–reaction condition regions that achieved over 2.5-fold improvements in turnover frequency and expanded the accessible regioselectivity range. Notably, Flex-Cat autonomously identified ligands exhibiting condition-programmable selectivity inversion (flexible ligands) across the same catalytic system. Data-driven structure–performance trends are extracted from the autonomously generated dataset, and top candidates are validated by translation to a 20 mL reactor format (10 × volume). Flex-Cat establishes a scalable and generalizable approach for autonomous catalyst development, linking discovery-scale experiments to process-relevant outcomes. The authors report Flex-Cat, a closed-loop autonomous catalysis platform that couples parallel miniaturized batch reactors for pressurized gas–liquid chemistry with a hierarchical, plate-constrained Bayesian optimization framework for mixed discrete (ligand identity) and continuous (process) variables.
Bennett et al. (Sat,) studied this question.