Development of QSARs/QSPRs to describe wastewater treatment and functionalized nanoparticles

Abstract

Endocrine disrupting compounds (EDCs), pharmaceuticals, personal care products, and their metabolites, are an increasing problem for industrial and municipal waste- and drinking- water treatment facilities. Many strategies exist for removal of harmful compounds, including physical separation by filtration or adsorption, transformation by chlorine or ozone oxidation, and combination uptake/transformation by biological systems. Testing all available protocols when a new problem compound is detected can be avoided if one or more options is predicted to perform poorly. To aid in testing minimization, QSARs with large domains of applicability are created for non-biological strategies (filtration, adsorption, and oxidation) utilizing datasets with large EDC populations. Within this work, a rarely-utilized inductive learning strategy, kernel multi-task latent analysis, is shown to produce more predictive models than single-task kernel partial least squares for membrane data. A web-based tool was created to allow public access to these models.

This dissertation utilizes traditional machine learning methods such as partial least squares and support vector machines to create quantitative structure-activity (or structure-property) relationships (QSARs/QSPRs) for materials and wastewater applications.

Polymer nanocomposites (PNCs) are a class of materials that shows utility in applications from aircraft hulls to medical prostheses. It is therefore desirable to predict the bulk properties of PNCs before their synthesis in order to avoid trial-and-error experimentation, which is expensive in both time and money. Prediction of glass transition temperature (T_g) of polymer composites using silica nanoparticles with organic functionalizations is completed using a three-tiered approach spanning atomic- micro- and macro- scales. Atomic-scale work includes QSPR prediction of the polar and dispersive components of surface tension for functionalized silica beads and polymers as separate systems. These surface tension components are then used to predict the the distribution of nanoparticles within the polymer matrix at the microscale, and the amount of interphase created by the distribution is used to predict the change in T_g relative to pure polymer. Prediction of dispersive and polar surface tension components for the functionalized silica particles is described in depth in this dissertation.