The development of quantitative structure-activity relationship models for physical property and biological activity prediction of organic compounds

The development of quantitative structure-property relationships (QSPRs) and quantitative structure-activity relationships (QSARs) is presented in this thesis. It is computationally expensive to find a direct relationship between chemical structure and a physical property or biological activity. To circumvent this, QSPR/QSAR uses an inductive approach to indirectly link structure and activity using numerical descriptors. The ultimate goal of QSPR/QSAR techniques is to construct mathematical models which can predict the physical properties or biological activities of compounds based on features of their chemical structure alone. Along with quantitative prediction of activity, classification models (binary QSARs) can also be utilized to categorize compounds into classes or groups of varying bioactivity.; First, the methods and procedures used to build QSPR/QSAR models are presented. The chemical structure of each compound is encoded via the calculation of a large pool of numerical descriptors.; The first application study involves the prediction of glass transition temperatures using the monomer and repeat unit structures of amorphous polymers.; Secondly, a set of carbonic anhydrase inhibitors is used to construct several QSAR and classification models for drug potency. The best QSAR models for each isozyme (CA I, CA II, and CA IV) of carbonic anhydrase were found using computational neural networks while k-nearest neighbor was used to solve two-class and three-class problems for inhibitors of the CA IV isozyme.; The third application study focuses on the development of QSAR and classification models to predict drug potency and selectivity for inhibitors of the dihydrofolate reductase enzyme. A data set of 345 inhibitors was used to build models for three types of dihydrofolate reductase: Pneumocystis carinii, Toxoplasma gondii, and rat liver.; The fourth application chapter involves the development of binary QSAR models to classify a data set of 334 aromatic and secondary amine compounds as genotoxic or nongenotoxic based on information calculated solely from chemical structure.; The development of a new suite of hydrophobic surface area (HSA) descriptors is the focus of Chapter 7.; The fifth and final application study involves the development of classification models for mutagenicity assessment using Ames test data with two diverse data sets of organic compounds. The models are developed for two strains of Salmonella typhimurium (TA100 and TA98) using 280 and 210 compounds, respectively. (Abstract shortened by UMI.)...