The determination of descriptors for catalytic systems in machine learning models using kinetic experimental data

The problem of selection and determining the values of descriptors for the properties of chemical reactions components in mathematical models for chemical processes is one of the essential ones when creating machine learning (ML) models used to describe and predict the functioning patterns of chemical systems. Current practice in the field mainly involves the use as the descriptors physical and chemical characteristics of the components of reaction systems (ionic radii, bond lengths, energies, and other parameters related to the structure and properties of specific molecules or particles) determined experimentally or by calculation. This work presents the results of the predicting of the integral kinetic dependences, as well as approaches to determine the values of descriptors for characterizing the properties of a set of simple palladium catalyst precursors when used in the Suzuki–Miyaura reaction. The problem stated has been solved by creating the ML models that take into account experimental kinetic data. The descriptors obtained as a result of training the models make it possible to satisfactorily describe the kinetic patterns of the Suzuki–Miyaura reaction with aryl chlorides under the so-called “ligand-free” catalytic conditions possessing higher sensitivity of the reaction to small changes in the conditions.