Helsingin yliopisto, Matemaattis-luonnontieteellinen tiedekuntaUniversity of Helsinki, Faculty of ScienceHelsingfors universitet, Matematisk-naturvetenskapliga fakultetenLempinen, Antti2023-09-132023-09-132023URN:NBN:fi:hulib-202309073725http://hdl.handle.net/10138/565055Development of machine learning models for reaction design has garnered growing interest. Notably, the benefits of predictive models include the elimination of trial and error in selecting suitable reaction conditions. In addition, mechanistic insight may be gained to help rational catalyst design. The aim of this study is to develop modeling and parametrization method for transition metal complexes which would enable the combined parametrization of mono- and bidentate ligands for the first time. Performance of novel parametrization method is demonstrated through a ligand classification model for Suzuki–Miyaura reaction. In this literature review, an overview of the computational catalyst modeling and performance prediction is presented. The history of physical and computational chemistry is reviewed, ranging from early structure-property relationships and linear models to modern physical organic density functional theory (DFT) parameters and machine learning models. The necessary theoretical background for reaction modeling is presented in terms of transition state theory, which can be used to model selectivity or reaction rate. Additionally, the reaction yield is discussed as typical performance score for reaction modeling. Furthermore, the initial structures of typical transition metal complexes used for modeling are presented, along with the methods employed for structure optimization and ligand parametrization. By utilizing this background, the evolution of modeling methods from linear free energy relationships to machine learning is discussed. Additionally, modern classification methods for catalyst design are reviewed. Finally, the mechanistic details of Suzuki–Miyaura reaction are explored to justify the modeling methodologies in the experimental part. In the experimental study, a novel method to combine parametrization for mono- and bidentate ligands was successfully invented. As a proof of concept, performance classification of ligands for Suzuki– Miyaura reaction was conducted. Suzuki–Miyaura reaction was chosen as the model reaction due to wide data availability. Initial transition metal complexes were built according to ligation state of nickel. The initial structures always included two carbonyl ligands and either one bidentate ligand, or two monodentate ligands, or one bulky monodentate ligand. The structures were optimized with semiempirical quantum mechanical method xTB and parametrized using a newly developed in-house parametrization method. Calculated parameters include global and local steric and electronic descriptors, and local geometric descriptors. Classification models were built with selected training data. Models were used to predict the performance of new ligands. Successful results were achieved, and ligands that provided better yields were identified. In addition, it was discovered that model reaction proceeds without the presence of phosphine ligand if superbase is used as a base. Nitrogen-containing bases were screened, additional active superbases were found, and correlating descriptor with activity was detected. Based on results, more active superbases were designed.engModeling, parametrization, classification, Suzuki–Miyaura reaction, ligand, catalyst, superbasepro gradu -tutkielmatKemiaChemistryKemiKemian ja molekyylitieteiden maisteriohjelmaMaster's Programme in Chemistry and Molecular SciencesMagisterprogrammet i kemi och molekylära vetenskaper