Advances in Feature Selection in One- and Two-Dimensional Gas Chromatography with Mass Spectrometry

Abstract

One- and two-dimensional gas chromatography coupled with mass spectrometry provides an enormous amount of quantitative data describing the chemical composition of complex samples. Besides quantification and identification of analytes, common analysis goals include classifying samples or predicting sample properties based upon the chemical information contained in the chromatographic data. The chemometric modeling techniques used to accomplish these goals often benefit from the removal of redundant or irrelevant chromatographic variables, which is achieved by feature selection. This dissertation presents several research studies detailing advances in and applications of feature selection applied to one- and two-dimensional gas chromatography with mass spectrometric detection. The two-dimensional mass cluster method was evaluated as a peak detection algorithm using simulations of gas chromatography coupled with time-of-flight mass spectrometry (GC-TOFMS) data under varying sample and separation complexity. An unsupervised feature selection method based on variance thresholding was applied to simulated GC-MS chromatograms and a previously studied yeast metabolome dataset. A successful application of partial least squares (PLS) regression analysis to comprehensive two-dimensional gas chromatography with time-of-flight mass spectrometry (GCÃ GC-TOFMS) for the prediction of bulk physical properties of kerosene-based fuels is included to demonstrate a case where feature selection was not required. Finally, supervised feature selection was implemented on GCÃ GC-TOFMS data of rocket fuels to aid in the prediction of fuel thermal integrity by PLS.