Effect of Solvents on Proline Modified at the Secondary Sphere: A Multivariate Exploration

Abstract

The critical influence of solvent effects on proline-catalyzed aldol reactions has been extensively described. Herein, we apply multivariate regression strategies to probe the influence of different solvents on an aldol reaction catalyzed by proline modified at its secondary sphere with boronic acids. In this system, both in situ binding of the boronic acid to proline and the outcome of the aldol reaction are impacted by the solvent-controlled microenvironment. Thus, with the aim of uncovering mechanistic insight and an ancillary aim of identifying methodological improvements, we designed a set of experiments, spanning 15 boronic acids in five different solvents. Based on hypothesized intermediates or interactions that could be responsible for the selectivity in these reactions, we proposed several structural configurations for the library of boronic acids. Subsequently, we compared the statistical models correlating the outcome of the reaction in different solvents with molecular descriptors produced for each of these proposed configurations. The models allude to the importance of different interactions in controlling selectivity in each of the studied solvents. As a proof-of-concept for the practicality of our approach, the models in chloroform ultimately led to lowering the ketone loading to only two equivalents while retaining excellent yield and enantio- and diastereo-selectivity.

Mechanistic Inference from Statistical Models at Different Data-Size Regimes

“…201,202 With this in mind, we speculated that the modularity of our in situ modified proline−boronic acid catalytic system is ideally suited for evaluating the influence of different solvents on aldol reactions. 193 We also anticipated that the variance incorporated in our data set design would serve two complementary purposes: guiding a mechanistic investigation and propelling the identification of an improved methodology. Thus, the data set was designed on the basis of a systematic variation of boronic acids in different common solvents (Figure 7A).…”

“…203−218 Pioneering work by the Fey group established libraries mostly specializing in phosphine ligands, motivated by their prevalence in cross-coupling reactions. 86,87,[193][194][195][196][197]200 More recently, a new library of monodentate phosphines, extended by ML techniques, was reported by the Aspuru-Guzik and Sigman groups. 211 The library is incorporated in an online platform named Kraken, and 200 descriptors are available for each ligand in the database.…”

“…Models incorporating these parameters can then be further investigated in order to raise nuanced mechanistic hypotheses that can then be tested experimentally. Our group has been focused mainly on small data sets (up to 75 experimental data points) to unveil mechanistic details of reactions catalyzed by second-sphere-modified organocatalysts. − In 2019, we reported that the catalytic activity of N-heterocyclic carbene (NHC) triazolium salts bearing a hydroxy handle could be fine tuned by the in situ addition of boronic acids as second-sphere modifiers. , …”

“…However, a drawback of our method was the use of 18 equiv of neat cyclopentanone with respect to the aldehyde as the reaction medium. Although organocatalyzed aldol reactions are highly sensitive to the reaction medium, studies aimed at understanding solvent effects in this reaction are rare. , With this in mind, we speculated that the modularity of our in situ modified proline–boronic acid catalytic system is ideally suited for evaluating the influence of different solvents on aldol reactions . We also anticipated that the variance incorporated in our data set design would serve two complementary purposes: guiding a mechanistic investigation and propelling the identification of an improved methodology.…”

“…Beyond trend evaluation, another powerful visualization technique for the identification of mechanism breaks are cutoff values in scatter plots. To streamline this approach, one can use computational ligand libraries that function as repositories of molecular parameters covering a broad chemical space. − Pioneering work by the Fey group established libraries mostly specializing in phosphine ligands, motivated by their prevalence in cross-coupling reactions. ,,− , More recently, a new library of monodentate phosphines, extended by ML techniques, was reported by the Aspuru-Guzik and Sigman groups . The library is incorporated in an online platform named Kraken, and 200 descriptors are available for each ligand in the database.…”