Data-Driven Analysis of Reactions Catalyzed by [CoCp*(CO)I₂]

Abstract

The combination of the [Cp*Co(CO)I 2 ] catalyst with various additives and solvents at different temperatures for different durations has led to a considerable increase in the amount of research regarding the catalysis of C−H functionalization via high-valent Co, owing to the unique reactivity of the catalyst. Our study presents, for the first time, a data-driven approach toward reported reactions catalyzed using [Cp*Co-(CO)I 2 ]. We analyze distinct solvents, additives, reaction times, temperatures, and directing groups with their associated mean yields, which accelerates the screening process in the laboratory to yield favorable combinations. We demonstrate the effects of reaction conditions and directing groups on the reaction performance by integrating data from the reaction. Additionally, we examine the reactions based on their yields using principal component analysis and the t-stochastic neighbor embedding method. These exploratory investigations may be very useful for synthetic chemists as a starting point in catalytic studies.

Machine Learning Yield Prediction from NiCOlit, a Small-Size Literature Data Set of Nickel Catalyzed C–O Couplings

“…The coverage of the substrate-coupling partners combinations is sparse and strongly biased toward a few reactions such as the Kumada coupling 20 (RMgX + ArOCH 3 ), which is the most investigated with 243 reported reactions. This datadriven analysis 21 of nickel-catalyzed couplings shows that published reaction data seems to focus on specific combinations. We were surprised to see that reactions were clustered not by reaction variables (for example, coupling partner) but almost perfectly by publications (Figure 1C).…”

Machine Learning-Driven Classification of Urease Inhibitors Leveraging Physicochemical Properties as Effective Filter Criteria

“…CADD is a collection of diverse computational techniques and resources, comprising compound databases, molecular simulations, structure-and ligand-based virtual screenings (VS), hit and lead optimization, quantitative structure-activity relationship (QSAR), among many others. The integration of various ML algorithms into the CADD process has greatly benefited pharmaceutical companies and academic research as ML provides them with innovative and efficient ways in every stage of the CADD process [44,45] and other branches of chemistry [46][47][48]. In ML, there exist two primary categories of algorithms: supervised learning and unsupervised learning.…”

Machine Learning Prediction of High-Yield Cobalt- and Nickel-Catalyzed Borylations

“…Some works are focused on reaction data extraction into a scheme consistent with databasing efforts like Reaxys (reaxys.com) and SciFinder n (scifinder-n.cas.org). However, the yield prediction problem is markedly different because, in addition to the structural features of reactants, products, and reagents, it entails “human” factors, − such as the chemist’s subjective preference for certain protocols or even the current availability of chemicals in the lab . Finally, reaction data taken from research articles tends to be biased, reporting only the best yields, and ML approaches for yield prediction generally have lower performances than those where HTE data is used. , …”