Towards Data‐Driven Design of Asymmetric Hydrogenation of Olefins: Database and Hierarchical Learning

Abstract

Asymmetric hydrogenation of olefins is one of the most powerfula symmetric transformations in molecular synthesis.A lthough several privileged catalyst scaffolds are available,t he catalyst development for asymmetric hydrogenation is still atime-and resource-consuming process due to the lacko fp redictive catalyst design strategy.T argeting the data-driven design of asymmetric catalysis,weherein report the development of as tandardized database that contains the detailed information of over 12000 literature asymmetric hydrogenations of olefins.T his database provides av aluable platform for the machine learning applications in asymmetric catalysis.B ased on this database,w ed eveloped ah ierarchical learning approach to achieve predictive machine leaning model using only dozenso fe nantioselectivity data with the target olefin, which offers au seful solution for the few-shot learning problem and will facilitate the reaction optimization with new olefin substrate in catalysis screening.

Genetic Optimization of Homogeneous Catalysts

“…Thanks to its hierarchical structure, it is compatible with diverse structural representations (e. g., SMILES, 3D structures), genetic operations and fitness functions. Additional functionalities, including ML‐based acceleration, [44–50] can also be conveniently deployed for the fitness evaluation. While NaviCatGA, as presented here, is a core component of inverse design efforts in catalysis, it also constitutes a powerful stand‐alone program for general optimization problems.…”

“…based acceleration, [44][45][46][47][48][49][50] can also be conveniently deployed for the fitness evaluation. While NaviCatGA, as presented here, is a core component of inverse design efforts in catalysis, it also constitutes a powerful stand-alone program for general optimization problems.…”

Bridging Chemical Knowledge and Machine Learning for Performance Prediction of Organic Synthesis

“…We recently built a database of asymmetric hydrogenation of olefins (12619 enantioselectivities) based on experimentation literature between the years 2000 and 2020. [45] In addition to the literature data, the reaction data schemes from US patents were extracted as the USPTO reaction database via text-mining techniques by NextMove. [46,47] However, it is noteworthy that, based on a recent study, there may be some inherent potential problems in the data source.…”

“…Prime examples of this strategy include Doyle's database of Ullman–Goldberg/Buchwald–Hartwig cross‐couplings (4140 reaction yields) [18] and Denmark's database of asymmetric imine addition (1075 enantioselectivities), [20] which have now been widely applied as benchmark databases for ML of synthesis/catalysis performance. We recently built a database of asymmetric hydrogenation of olefins (12619 enantioselectivities) based on experimentation literature between the years 2000 and 2020 [45] . In addition to the literature data, the reaction data schemes from US patents were extracted as the USPTO reaction database via text‐mining techniques by NextMove [46,47] .…”

Mechanistic Inference from Statistical Models at Different Data-Size Regimes

“…To ensure statistical relevance, the input data should cover a wide range of outcomes, assuring that they properly represent the system under investigation. This can be a challenge when the synthetic chemistry literature is mined, as it is not common practice to report results of failed, unselective, or low-yielding reactions. ,,− Consequently, when a mechanistic approach is adopted on the basis of statistics, it is often the case that data mining the literature is not enough and data sets need to be augmented experimentally. In 2018, the Doyle group started investigating HTE data sets for the predictions of reaction yields .…”