Heuristic Data-Driven Evolution of Synergistic Cobalt (IV)-Enamine Catalysis

Discovering new chemical reactions is often a slow, intuition-driven process, limited by the vastness of chemical space and the challenges of pinpointing optimal conditions. Here, we introduce a data-driven strategy that integrates expert intuition with AI-guided exploration, enabling rapid identification of effective reaction conditions with minimal experimentation. Applying this approach, we establish an Co(IV)-enamine catalytic system that enables a new strategy on the α-polarity inversion of carbonyl compounds. A Bayesian optimization framework systematically explored over one million reaction conditions, converging on an optimized system within just 63 experiments. This strategy identified superior reaction conditions with minimal screening, uncovered unexpected reactivity patterns, and expanded the substrate scope through AI-guided clustering analysis. By integrating human expertise with AI-driven exploration, this workflow establishes a scalable and generalizable paradigm for reaction discovery, offering a powerful tool for more efficient and targeted development of new catalytic transformations.