A Data Science Approach to Organic Reaction Discovery and Development

点击翻译按钮获取中文摘要

Reaction development is focused on the identification of optimal conditions that facilitates the conversion of starting materials to a desired product. The application of conditions from closely-related reactions (e.g. those involving similar substrates) to the target transformation largely drives the current process of method development. Unfortunately, this approach often fails owing to subtle differences in reaction requirements. Therefore, the optimization process continues to be an empirical endeavor. Our program plans to address this challenge by developing new data science approaches capable of predicting and interpreting reaction outcomes. This work would have three key impacts in chemical synthesis: 1) streamlining the costly process of reaction optimization, 2) enable reaction application to complex substrates. And 3) as the data driven tools utilize physical organic methods to define molecules mathematically, the resulting correlation can be interpreted to provide insights into how catalysts/substrates interact. In our approach, DFT-derived parameter sets describing the structural features of the reaction components are related to experimental outputs. The resulting equation can be deployed to predict experimental outcomes when these components are altered. The reactions and catalysts under study will range significantly in structure and application, but the general goal is to understand the interactions responsible for effective catalysis and to develop new data-driven tools that will facilitate reaction design. This proposal will be driven by the investigation of two critical new ideas. The first is to build data science workflows with enhanced prediction tools for reaction screening to enable rapid identification of optimal conditions for catalysis. The second asks questions pertaining to the generality of selectivity determining interactions in mechanistically disparate reactions. Are the non-covalent interactions governing the enantioselectivity outcomes with similar catalyst or substrate types fundamentally the same? Answering this important question will have long-ranging implications on reaction modeling and allow us to solve the following specific problems: 1) How far "out-of-sample" can the model(s) extrapolate? 2) How can different catalyst chemotypes be included in the same correlation? 3) Can general mechanistic models be developed? In sum, we propose a research program that aims to develop and advance predictive strategies to solve key problems in organic synthesis. Ultimately, such tools will enhance our ability to apply known reactions to complex targets and develop new methods, immediately impacting how one performs experiments in the laboratory. This research will provide unique training opportunities for the next generation of modern organic chemists and will lead to benefits for Canadians by developing new methods for streamlining the reaction optimization that could be implemented in industrial settings.