Development of a Computer-Guided Workflow for Catalyst Optimization. Descriptor Validation, Subset Selection, and Training Set Analysis

Abstract

Modern, enantioselective catalyst development is driven largely by empiricism. Although this approach has fostered the introduction of most of the existing synthetic methods, it is inherently limited by the skill, creativity, and chemical intuition of the practitioner. Herein, we present a complementary approach to catalyst optimization in which statistical methods are used at each stage to streamline development. To construct the optimization informatics workflow, a number of critical components had to be subjected to rigorous validation. First, the critically important molecular descriptors were validated in two case studies to establish the importance of conformation-dependent molecular representations. Next, with a large data set available, it was possible to investigate the amount of data necessary to make predictive models with different modeling methods. Given the commercial availability of many catalyst structures, it was possible to compare models generated with algorithmically selected training sets and commercially available training sets. Finally, the augmentation of limited data sets is demonstrated in a method informed by unsupervised learning to restore the accuracy of the generated models.

Design and Application of a Screening Set for Monophosphine Lig-ands in Metal Catalysis

“…The specific workflow for the selection of a diverse molecule set adopted herein (Figure 2) represents just an example of a more general procedure where different choices can be made at every step, depending on the application. Indeed, others have used similar procedures to generate diverse sets of molecules in unique contexts, 7,9,[11][12][13][14][15][16] including phosphorus ligand sets. 8,[17][18][19] First, selecting a set of ligands that covers the entire range of ligand properties requires a means to quantify this range in advance, thus defining the search space.…”

Computational Methods for Training Set Selection and Error Assessment Applied to Catalyst Design: Guidelines for Deciding Which Reactions to Run First and Which to Run Next

“…Gasteiger charges 6 for each structure was then calculated using openbabel. 7 AEIF and ASO descriptors (previously published by our laboratory) 8 were then calculated for each amino alcohol. To represent the identifty of the metal, NBO charges of the precatalysts, highest occupied metal orbital, lowest unoccupied metal orbital, and sterimol B5 and L parameters using all atoms from the arene or lewis basic ligand for the precatalyst structures were tabulated.…”

A Conformer‐Dependent, Quantitative Quadrant Model

“…Our own laboratory has developed an informatics-driven protocol using machine learning to predict optimal catalysts with unoptimized reaction results using 3D molecular descriptors. [43,44] The quadrant diagram is a classic example of a qualitative tool for rationalizing stereoselectivity in enantioselective transformations. This analysis, originally developed by Knowles for enantioselective hydrogenation, [45] has been used post facto to rationalize the stereoselectivity of catalysts.…”

“…These approaches do not rely on human intuition to identify catalyst substructures important for stereoinduction since they take into account the entire structure of the catalyst. Our own laboratory has developed an informatics‐driven protocol using machine learning to predict optimal catalysts with unoptimized reaction results using 3D molecular descriptors [43,44] …”