Bioinformatics for mass spectrometry. Novel statistical algorithms

Mass spectrometry is a technique to determine the molecular content of samples derived from human, animal and plant. It can be used for the discovery of biomarkers – molecules which discriminate between, for example, a healthy and a diseased organism. Due to physical and chemical phenomena, a given molecule species can give rise to a series of interrelated peaks in a mass spectrum. Therefore, a realistic sample can result in a very complex spectrum with many interrelated and overlapping peaks. Today’s methods provide biased/erroneous predictions of the molecular masses and abundances in the measured samples when analyzing (complex) spectra. This thesis outlines the sources of variation which can affect a mass spectrum. Based on this, novel statistical models and methods are derived, which are robust to noise and very suitable for the detection and unbiased quantification of (severely overlapping) peaks. After computationally linking interrelated peaks, peaks are shifted to their correct positions in the spectrum, and eventually combined into a much smaller list of ‘true’ molecules. This enhances the ‘statistical power’ and thus the chances on finding biomarkers. In addition, methods are developed for the analysis of mass spectrometry time series data. These methods can help to monitor the effects of surgical, nutritional and pharmacological interventions, and to predict patient recovery.