Desenvolvimento de novas metodologias de acoplamento C-C e/ou C-N: mesclando ciência de dados e catálise metálica

Abstract

The approach of statistical methods capable of accurately predicting the relationship between structure and reactivity represents a major impact on the development of reactions. Recently, machine learning tools have been guided and applied in synthesis design. In the context of the work described here, these methods provide rapid information and relevant estimates about the structure and respective activity of substrates that are summarized in structural descriptors that influence the desired activity. Here we list the main results obtained in the development and use of substrate parameterization in new methodologies. In the context of nucleopalladation reactions, we highlight Wacker-type reactions that employ carbonylation reactions of non-activated double bonds by CO capture in a palladium- catalyzed process. We present two new methodologies, one of them to obtain pyrazoline ester derivatives tolerant to different substitutions in strategic positions of the starting material as demonstrated by the scoping study. As well, a new methodology for obtaining pyrazolines with a ketone moiety using boronic acids which allowed access to structural diversity ketone derivatives not previously described guided by a virtual library of boronic acids. Analysis of the electronic and steric factors into the reactivity was fundamental for understanding the nucleophilicity necessary for boronic acids in the transmetalation step. In sequence, we report our efforts to integrate data science and computational chemistry tools to guide, predict and explain the reactivity of persistent radicals generated in the reduction of cyano-arenes in the electrophilic cross-coupling between cyano(hetero)arenes and alkyl halides. The selection of substrates was made from the construction of a virtual library of cyanoarenes via projection of the chemical space by UMAP based on the dimensionality reduction of DFT level physicochemical parameters, ensuring structural diversity in relation to the chemical space. A predictive univariate model could be generated by correlating an electronic parameter with yield.