Towards the Computer-Aided Molecular Design of Reactants and Products
Type of DegreePhD Dissertation
MetadataShow full item record
Many chemical processes generate products with properties of interest to businesses and end consumers (B&EC), using reactions. Thus, a need for quantitative modeling of physico-chemical properties and molecular design (MD) in reactive systems has arisen. Quantitative modeling precedes MD and is useful in relating chemical structure to properties of interest. Using quantitative models, a chemical’s properties can be systematically varied by varying its structure. This structure variation, without sole reliance on intuition, assists in exploring a large portion of the chemical space. When the rising prowess of computers is tapped, such an exploration is termed as computer-aided molecular design (CAMD). CAMD of products of reactions is thus beneficial since the demands of B&EC can be met efficiently. Since products originate from reactants, CAMD of products will also lead to the CAMD of reactants. While CAMD of solvents and catalysts has received significant attention, there is a paucity of CAMD algorithms that design reactants and products. To address this paucity, CAMD of reactants and products in three scenarios has been explored in this work. In the first scenario, only the products’ respective dominant properties are optimized, given a set of property constraints. In the second scenario, properties that are dependent on the structures of both reactants and products are optimized. Unlike the first scenario, both reactants and products are subject to property constraints. In the third scenario, each reactant and product’s respective dominant property is optimized. Like the second scenario, both reactants and products are subject to property constraints. Our CAMD methodologies incorporate property models with a variety of molecular descriptors using signature descriptors, which are molecular building blocks. In order to generate feasible structures, previously developed structural constraints have been improved. Since the structures of reactants and products are related, relationships have been derived between them using signature descriptors. To demonstrate the efficacy of the developed CAMD methodologies, a case study has been solved for each scenario. Additionally, for CAMD of reactants, products and solvents for reaction rate optimization, we compare promising ensemble learning algorithms’ abilities to model reaction rate constant in terms of structures of reactants and solvents. We assessed decision tree-based ensemble methods’ abilities to model the Diels-Alder reaction’s rate constant in a case study.