| This thesis develops additional validation steps in the laboratory workflow for the search program peptide assignments in shotgun proteomics. The current search programs, algorithms, and process models are not sufficiently accurate or powerful for this important task. Proteolytic cleavage patterns are examined for unusual amino acid sequences near the cleavage site using statistical sequence patterns found in our high confidence peptide datasets. This characterization allows for the rejection of a percentage of invalid search engine assignments. The reverse phase chromatography step that separates peptides based on their hydrophobicity is also examined; the retention time patterns are modeled for compositions of the amino acids that are compatible with the observed elution. Multilinear regression and neural networks are applied and compared to each other for predicting retention, with investigations of the optimal methods and architectures for the discrimination task. Improvements in the false positive rate and relaxations to improve the false negative rate are shown for MudPIT sample runs. The work herein at this time is important because a more efficient and rapid assessment of the proteome, which is the complete set of expressed proteins over time, is crucial in understanding the biochemistry and molecular pathways of life processes and how to influence them pharmacologically to treat metabolic disorders. |
