Open Collections

Abstract

The pursuit of highly selective and broadly applicable catalytic transformations has driven decades of innovation in synthetic chemistry. However, the design of new catalytic systems and the a priori prediction of their performance remain elusive, relying heavily on empirical screening and chemical intuition. In recent years, the fields of machine learning and data science have offered powerful new paradigms for navigating such complex, high-dimensional problems. These data- driven approaches hold the promise of uncovering subtle structure-activity relationships that elude human intuition, leading to predictive models that pose to greatly accelerate catalyst design and reaction optimization. This thesis details the development and application of tailored machine learning frameworks to interrogate the fundamental principles governing selectivity in asymmetric catalysis, with a focus on providing actionable insights for experimental chemists. The first section of this thesis deals with expanding the use of techniques based on Linear Free Energy Relationships (LFERs), specifically demonstrating how these models can be leveraged by experimentalists to aid in target synthesis campaigns and how simple classifiers can be used to accelerate mechanistic interpretation. I further show that the identification of LFERs themselves can be a powerful approach to generate new mechanistic hypotheses from chemically heterogenous datasets. Next, I establish principles for optimal data collection in data-limited scenarios, demonstrating that a target-aware protocol is more efficient for building predictive models than maximizing dataset diversity. Finally, I propose a workflow built on unsupervised machine learning to quantify generality, the ability of catalysts to selectively catalyze diverse reactions. I show that this data-driven approach can streamline the identification of general catalysts, a process typically requiring decades of experimental effort, in addition to accelerating reaction optimization. Overall, this thesis aims to bridge the gap between data-driven methodologies and experimental insights, providing researchers with predictive tools to make catalyst and reaction design a more rational and efficient process.