Rationally designed catalysis for the enantioselective activation of alkenes

点击翻译按钮获取中文摘要

The development of a fast and robust screening platform using an automated synthesis robot and machine learning is the focus of our efforts this year, based on the original proposal.While traditional approaches to optimizing catalytic processes rely on inductive and qualitative assumptions drawn from screening data, recent developments in machine learning models offer a more quantitative evaluation of data. However, these models can be expensive due to the required quantum chemical calculations. To avoid these costs, 2D descriptors such as fragment counts or binary fingerprints, which represent general structural features, could be used. Although binary fingerprint descriptors are accessible and cost-effective, their predictive performance has been limited. To overcome this issue, we developed a machine learning model that employs fragment descriptors, fine-tuned for asymmetric catalysis and representing cyclic or polyaromatic hydrocarbons, which enabled efficient and robust virtual screening. Using training data with moderate selectivities, we designed and validated new catalysts that exhibit higher selectivities in a challenging asymmetric tetrahydropyran synthesis.The details of this work can be found in our publication in Angew. Chem. Int. Ed. (10.1002/anie.202218659).