Content area
Abstract
Technological advances in proteomics have shown great potential in detecting cancer at the earliest stages. One way is to use the time of flight mass spectroscopy to identify biomarkers, or early disease indicators related to the cancer. Pattern analysis of time of flight mass spectra data from blood and tissue samples gives great hope for the identification of potential biomarkers among the complex mixture of biological and chemical samples for the early cancer detection. One of the keys issues is the pre-processing of raw mass spectra data. A lot of challenges need to be addressed: unknown noise character associated with the large volume of data, high variability in the mass spectroscopy measurements, and poorly understood signal background and so on. This dissertation focuses on developing statistical algorithms and creating data mining tools for computationally improved signal processing for mass spectrometry data. I have introduced an advanced accurate estimate of the noise model and a half-supervised method of mass spectrum data processing which requires little knowledge about the data.
