Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI) is one of several proteomics technologies that can be used in biomarker discovery studies. Such studies often have the goal of finding protein markers that predict early onset of cancers such as cervical cancer. The reproducibility of SELDI has been shown to be an issue in the literature. There are numerous sources of error in a SELDI experiment starting with sample collection from patients to the signal processing steps used to estimate the protein mass and abundance values present in a sample.

This dissertation is concerned with all aspects of signal processing related to SELDI’s use in biomarker discovery projects. In chapter 2, we perform a comprehensive study of the most popular preprocessing algorithms available. Next, in chapter 3, we study the basic statistics of SELDI data acquisition. From here, we propose a quadratic variance measurement model for buffer+matrix only spectra. This model leads us to develop a modified Antoniadis-Sapatinas wavelet denoising algorithm that demonstrates superior performance when compared to MassSpecWavelet, one of the leading techniques for preprocessing SELDI data. In chapter 4, we show that the quadratic variance model 1) extends to real pooled cervical mucus QC data from a clinical study, 2) predicts behavior and reproducibility of peak heights, and 3) finds four times as many reproducible peaks as the vendor-supplied preprocessing programs.

The quadratic variance measurement model for SELDI data is fundamental and promises to lead to improved techniques for analyzing the data from clinical studies using this instrument.