Peter Ulintz, Ph.D.
08
Chair
Dissertation Title
Research Interest
The proteomics field is driven by the need to develop increasingly high-throughput methods for the identification and characterization of proteins. The overall goal of this research is to improve the success rate of modern high-throughput proteomics studies. The focus is on developing computational strategies for increasing the number of identifications as well as improving the ability to distinguish new forms of proteins and peptides. Several studies are presented, addressing different points in the proteomics analysis pipeline. At the most fundamental data analysis level, methods for using modern machine learning algorithms to improve the ability to distinguish correct from incorrect peptide identifications are presented. These techniques have the potential to minimize the need for manual curation of results, providing a significant increase in throughput in addition to increased identification confidence. Non-standard types of mass spectrometry data are being generated in specific contexts. Specifically, phosphoproteomics often involves the generation of MS3 spectra. These spectra alleviate problems associated with MS2 fragmentation of phosphopeptides, but utilizing the additional information contained in these spectra requires novel informatics. Several strategies for accommodating this additional information are presented. A statistical model is developed for translating the information contained in the coupling of consecutive MS2 and MS3 spectra into a more accurate peptide identification probability score. Also, methods for combining MS2 and MS3 data are explored. A newer mass spectrometry methodology useful for phosphoproteomics has recently been introduced as well, termed multistage activation (MSA). A comparative study of this and other methods is presented aimed at determining an optimal method for generating phosphopeptide identifications, focusing not only on data analysis techniques, but also on the mass spectrometry methodologies themselves. A dataset is presented from a differential study of a human cell line infected with the dengue virus. The study explores the complementarity of different fractionation methods in generating more unique protein identifications. A discussion of a statistical mixture model that utilizes relative quantification information to classify identified peptides into two categories based on their membrane topology is given in the final chapter. Finally, a comment on utilizing pI information to enrich for phosphopeptides is provided.