Data Science Enables the Development of a New Class of Chiral Phosphoric Acid Catalysts

In-silico-assisted derivatization of triarylboranes for the catalytic reductive functionalization of aniline-derived amino acids and peptides with H2

Cheminformatics-based machine learning (ML) has been employed to determine optimal reaction conditions, including catalyst structures, in the field of synthetic chemistry. However, such ML-focused strategies have remained largely unexplored in the context of catalytic molecular transformations using Lewis-acidic main-group elements, probably due to the absence of a candidate library and effective guidelines (parameters) for the prediction of the activity of main-group elements. Here, the construction of a triarylborane library and its application to an ML-assisted approach for the catalytic reductive alkylation of aniline-derived amino acids and C-terminal-protected peptides with aldehydes and H2 is reported. A combined theoretical and experimental approach identified the optimal borane, i.e., B(2,3,5,6-Cl4-C6H)(2,6-F2-3,5-(CF3)2-C6H)2, which exhibits remarkable functional-group compatibility toward aniline derivatives in the presence of 4-methyltetrahydropyran. The present catalytic system generates H2O as the sole byproduct.

Computationally Guided Ligand Discovery from Compound Libraries and Discovery of a New Class of Ligands for Ni-Catalyzed Cross-Electrophile Coupling of Challenging Quinoline Halides

Although screening technology has heavily impacted the fields of metal catalysis and drug discovery, its application to the discovery of new catalyst classes has been limited. The diversity of on- and off-cycle pathways, combined with incomplete mechanistic understanding, means that screens of potential new ligands have thus far been guided by intuitive analysis of the metal binding potential. This has resulted in the discovery of new classes of ligands, but the low hit rates have limited the use of this strategy because large screens require considerable cost and effort. Here, we demonstrate a method to identify promising screening directions via simple and scalable computational and linear regression tools that leads to a substantial improvement in hit rate, enabling the use of smaller screens to find new ligands. The application of this approach to a particular example of Ni-catalyzed cross-electrophile coupling of aryl halides with alkyl halides revealed a previously overlooked trend: reactions with more electron-poor amidine ligands result in a higher yield. Focused screens utilizing this trend were more successful than serendipity-based screening and led to the discovery of two new types of ligands, pyridyl oxadiazoles and pyridyl oximes. These ligands are especially effective for couplings of bromo- and chloroquinolines and isoquinolines, where they are now the state of the art. The simplicity of these models with parameters derived from metal-free ligand structures should make this approach scalable and widely accessible.

A genetic optimization strategy with generality in asymmetric organocatalysis as a primary target

A catalyst possessing a broad substrate scope, in terms of both turnover and enantioselectivity, is sometimes called "general". Despite their great utility in asymmetric synthesis, truly general catalysts are difficult or expensive to discover via traditional high-throughput screening and are, therefore, rare. Existing computational tools accelerate the evaluation of reaction conditions from a pre-defined set of experiments to identify the most general ones, but cannot generate entirely new catalysts with enhanced substrate breadth. For these reasons, we report an inverse design strategy based on the open-source genetic algorithm NaviCatGA and on the OSCAR database of organocatalysts to simultaneously probe the catalyst and substrate scope and optimize generality as a primary target. We apply this strategy to the Pictet-Spengler condensation, for which we curate a database of 820 reactions, used to train statistical models of selectivity and activity. Starting from OSCAR, we define a combinatorial space of millions of catalyst possibilities, and perform evolutionary experiments on a diverse substrate scope that is representative of the whole chemical space of tetrahydro-β-carboline products. While privileged catalysts emerge, we show how genetic optimization can address the broader question of generality in asymmetric synthesis, extracting structure-performance relationships from the challenging areas of chemical space.

Dataset Design for Building Models of Chemical Reactivity

Models can codify our understanding of chemical reactivity and serve a useful purpose in the development of new synthetic processes via, for example, evaluating hypothetical reaction conditions or in silico substrate tolerance. Perhaps the most determining factor is the composition of the training data and whether it is sufficient to train a model that can make accurate predictions over the full domain of interest. Here, we discuss the design of reaction datasets in ways that are conducive to data-driven modeling, emphasizing the idea that training set diversity and model generalizability rely on the choice of molecular or reaction representation. We additionally discuss the experimental constraints associated with generating common types of chemistry datasets and how these considerations should influence dataset design and model building.

Data science enabled discovery of a highly soluble 2,2'-bipyrimidine anolyte for application in a flow battery

Development of non-aqueous redox flow batteries as a viable energy storage solution relies upon the identification of soluble charge carriers capable of storing large amounts of energy over extended time periods. A combination of metrics including number of electrons stored per molecule, redox potential, stability, and solubility of the charge carrier impact performance. In this context, we recently reported a 2,2'-bipyrimidine charge carrier that stores two electrons per molecule with reduction near -2.0 V vs. Fc/Fc+ and high stability. However, these first-generation derivatives showed a modest solubility of 0.17 M (0.34 M e-). Seeking to improve solubility without sacrificing stability, we harnessed the synthetic modularity of this scaffold to design a library of sixteen candidates. Using computed molecular descriptors and a single node decision tree, we found that minimization of the solvent accessible surface area (SASA) can be used to predict derivatives with enhanced solubility. This parameter was used in combination with a heatmap describing stability to de-risk a virtual screen that ultimately identified a 2,2'-bipyrimidine with significantly increased solubility and good stability metrics in the reduced states. This molecule was paired with a cyclopropenium catholyte in a prototype all-organic redox flow battery, achieving a cell potential up to 3 V.

Machine Learning Strategies for Reaction Development: Toward the Low-Data Limit

Machine learning models are increasingly being utilized to predict outcomes of organic chemical reactions. A large amount of reaction data is used to train these models, which is in stark contrast to how expert chemists discover and develop new reactions by leveraging information from a small number of relevant transformations. Transfer learning and active learning are two strategies that can operate in low-data situations, which may help fill this gap and promote the use of machine learning for tackling real-world challenges in organic synthesis. This Perspective introduces active and transfer learning and connects these to potential opportunities and directions for further research, especially in the area of prospective development of chemical transformations.