Accelerating Discovery by Merging Kinetics with Data Science Methods in Catalytic Organic Reactions - Project Summary/Abstract Over the past three decades, the Blackmond laboratory has pioneered research in kinetic-assisted mechanistic analysis of organic catalytic reactions. The data-rich methodology of Reaction Progress Kinetic Analysis (RPKA), employing in-situ reaction monitoring and kinetic modeling of temporal reaction profiles, has proven to be an innovative tool in organic synthesis and catalysis. Our work has helped to streamline pharmaceutical reaction process understanding as well as to provide fundamental mechanistic insights that contribute to catalyst design and development. Looking forward at the future landscape for research in synthetic organic chemistry, the significant potential for data-rich science focuses on artificial intelligence and machine learning (AI/ML) methods as an important theme in both academic and industrial research, fueling the promise of accelerated discovery. The experience and accomplishments of the Blackmond laboratory positions this group to make significant advances over the next five years in the study of complex catalytic networks, sometimes referred to as “systems chemistry.” Coupled catalytic cycles operating together may exhibit emergent behavior that cannot be predicted by a reductive approach studying single reaction steps in isolation. Key to our proposal is that we offer an approach that is orthogonal to current data science work in organic synthesis. Most investigations of organic catalytic reactions extract a single piece of information from each reaction – an endpoint yield or selectivity – which may be sufficient to characterize simple catalytic cycles; however, structural and electronic molecular descriptors applied to catalyst systems of higher complexity may by themselves fail to produce accurate predictive models of the network behavior. To address this challenge, we plan to incorporate dynamic information about complex reaction networks by incorporating images of kinetic profiles as inputs into AI/ML models. Capturing this comprehensive temporal information – the “narrative” of an entire reaction sequence – in AI/ML models may provide mechanistic insights that will uncover and exploit emergent chemical behavior and aid in the identification of new reaction and catalyst combinations. Two general classes of reaction networks will be developed as proof of concept of this systems-based approach: i) asymmetric catalytic cascade reaction networks involving sequential catalytic cycles, in which the products of one cycle become the reactants in the next, much like metabolic cycles in biology; and ii) synergistic multi-catalyst networks incorporating several catalytic cycles that are interconnected as cogs in a process, where intrinsic reactivity of each catalyst must be balanced with the other cycles in the network. We seek to implement predictive designs of novel reaction sequences by developing kinetics-based markers using the temporal reaction profile as a design tool. A key goal is to broaden the chemical space accessed by these models.
