Annotating proteomic mass spectrometry data for qualitative and quantitative experiments

This dissertation presents and evaluates algorithms that annotate mass spectrometry data from qualitative and quantitative experiments. The problems encountered when using mass spectrometry to predict the condition of samples (e.g. disease), and to determine the relative differences in protein expression from two samples, motivate the development of these algorithms.;This dissertation introduces algorithms that couple statistical learning and dynamic programming to annotate peaks from a mass spectrum into groups that correspond to the underlying chemistry of the studied samples. We demonstrate the utility of annotating these groups over the expert's annotations and other similar algorithms currently in the field. During this process the potential of using a "mostly" correct example set over a "near-miss" example set is shown to improve the annotation results. Using these annotated groups as features for classification, rather than just the individual peaks, improves the prediction accuracy for Prion disease and provides a better reflection upon the underlying chemistry.;Next, this dissertation derives and evaluates algorithms that annotate mass spectrometry data from quantitative experiments containing a metabolic label. The algorithm takes as input the annotated groups from the previous algorithm and subsequently matches the "light" and "heavy" pairs of molecules that have the same atomic formula, but are different in their atomic isotope content.;Finally, this dissertation compares and contrasts algorithms that calculate the relative quantitative ratios from the annotated isotopic matched pairs. It then presents a metric for measuring the annotation and quantitative calculations simultaneously.;Overall, this dissertation demonstrates the potential utility of statistical learning applied to the problems of qualitative and quantitative mass spectrometry.