| Description |
Multivariate linear regression (MLR) modeling is a data-driven strategy for investigating molecular reactivity to identify key mechanistic features of the reaction that influence reaction performance. Catalytic reactions are often multifaceted, which renders the reaction development challenging and conventionally requires synthesis of multiple analogs of the catalyst. In this dissertation, MLR models that quantitatively evaluate molecular descriptors as a function of various catalytic reaction outcomes (enantioselectivity, site selectivity, and turnover frequency) were utilized concurrent to reaction optimization, enabling prediction of an optimal ligand and providing mechanistic understanding of the reaction. In Chapter 2, an extensive library of molecular descriptors for aryl ring substituent effects was collected from simulated structures of ortho-, meta-, and para-substituted benzoic acids. Proximal and remote steric effects were accounted for in the obtained selectivity-MLR models investigating various arene substrates. The effect of the catalyst on the 1) enantioselectivity in an intermolecular dehydrogenative Heck arylation of indoles and alkenyl alcohols and 2) site selectivity in palladium-catalyzed 1,3-arylfluorination of chromenes and boronic acids in the presence of Selectfluor was evaluated in Chapter 3 by using molecular descriptors of [N,N]-bidentate ligands in MLR models. From a virtual screening library of ligand parameters, a new chiral pyridine oxazoline ligand was identified affording improved enantioselectivity in the Heck iv arylation. The mechanistic study of the arylfluorination reaction revealed that the site selectivity was influenced by the denticity and electronic nature of the ligand. To further explore catalyst reactivity, metathesis catalyst ligand parameters were utilized in MLR models described in Chapter 4 to quantitative evaluate 1) turnover frequency in self-metathesis of cis-nonene using silica-supported tungsten catalysts and 2) selectivity for ethenolysis of cis-cyclooctene using ruthenium N-heterocyclic carbene metathesis catalysts. The generation and utilization of a significant-sized library of simulated ligand parameters for MLR modeling provides a platform for accelerated catalyst development by virtual prediction of reaction outcome a priori. Harnessing the predictive power of MLR models is advantageous in reducing the synthetic effort invested on screening multiple analogs of a particular catalyst. Ultimately, the mechanistic rationale obtained from these parameterization studies can lead to the design of new catalysts and reaction optimization strategies. |
