AWWA WQTC71387

This paper describes the first step towards building quantitative structure-property relationship (QSPR) models to systematically select a group of representative micropollutants, which will serve as a training set to develop QSPR models for water treatment processes. A well developed optimized selection strategy was applied, which combined principal component analysis (PCA) and statistical experimental design. In this research, the initial dataset contained 183 micropollutants, mostly emerging contaminants, selected from the peer-reviewed literature. Each compound was characterized by 858 molecular descriptors (i.e. these are variables used in QSPR modeling). This resulted in a large complex multivariate dataset to which PCA was applied to summarize the information in the form of principal components. The first four principal components which captured 62.9% of the variation in the initial dataset were used to select representative compounds using a D-optimal onion design approach. Using this design, 22 substances were selected as structurally representative compounds which covered the chemical domain (meaning the chemical characteristics of all compounds) in a well-balanced manner and captured the majority of the information. The systematic selection approach employed here ensures that future QSPR models are applicable to a wide range of chemicals as long as their characteristics fall within the original chemical domain. Includes 15 references, tables, figure.