Alignment and Variable Selection Tools for Gas Chromatography -- Mass Spectrometry Data

Each individual chromatogram is typically very sparse, especially when coupled with a technique such as mass spectrometry. Only a small portion of the recorded data actually contains potentially useful information. Since the inclusion of irrelevant information will have detrimental effects upon the resulting chemometric model, variable selection is an essential step in pre-processing of raw chromatographic data. Presented here is a model evaluation metric named cluster resolution which was developed to, when used in conjunction with variable ranking metrics such ANOVA or selectivity ratio, guide the variable selection process.;Retention time shifts are another problem that makes application of chemometric techniques to chromatographic data more challenging. While numerous alignment algorithms are available, alignment remains a challenge when dealing with highly dissimilar samples. To solve this problem, deuterated alkane ladder-based alignment was developed.;Finally, the tools developed in the course of this study were applied to pre-processing of casework fire debris data for the purpose of classification of fire debris based on ignitable liquid content. This project was performed in collaboration with Royal Canadian Mounted Police. The developed tools allowed effective chromatographic alignment and variable selection, which permitted the construction of the first chemometric models able to reliably classify casework fire debris based on the presence or absence of gasoline.