Computer-aided development of quantitative structure-activity relationships and analysis of data from an artificial nose

Quantitative structure-property (or activity) relationships and the analysis of data generated from a device that mimics the mammalian olfactory system were both studied in this thesis. Quantitative structure-property (or activity) relationships (QSPRs or QSARs) are mathematical models that relate a compounds structure to a property or activity of interest. A successful QSAR (toxicity) and QSPR (aqueous solubility) were both designed by calculating descriptors that were mathematical representations of the structure of each compound, and then by using combinatorial optimization techniques to find the most information-rich subset of descriptors. A major focus of the research was the implementation of simulated annealing for optimal descriptor subset selection. Optimization algorithms were developed to select subsets based on the results of both multiple linear regression and computational neural networks.; The research also consisted of the analysis of data from an artificial nose. The mammalian olfactory system is known to possess both a broad-band response and remarkable sensitivity. It functions by use of a large array of cross-reactive receptor cells, rather than species-specific sensors. There have been many attempts to mimic this property in the design of vapor-sensing instruments, usually incorporating an array of chemical sensors.; In this work, data from an array of fiber-optic sensors provided input to pattern recognition techniques (feed-forward neural networks and K-nearest neighbor approach). The optical sensor array consisted of nineteen fiber optics coated with Nile Red immobilized in various polymer matrices. Responses consisted of the change in fluorescence with time resulting from the presentation of a vapor to the sensor array. Numerical descriptors calculated from these responses were then used to highlight important temporal features. Pattern recognition techniques were able to accurately identify and quantify each of the presented analytes. Three separate data sets were classified to approximately 90% accuracy, and the odorants of the first data set were also quantified to 97% accuracy.