Quantification and Simulation of Liquid Chromatography-Mass Spectrometry Data

Abstract

Computational mass spectrometry is a fast evolving field that has attracted increased attention over the last couple of years. The performance of software solutions determines the success of analysis to a great extent. New algorithms are required to reflect new experimental procedures and deal with new instrument generations. One essential component of algorithm development is the validation (as well as comparison) of software on a broad range of data sets. This requires a gold standard (or so-called ground truth), which is usually obtained by manual annotation of a real data set. Comprehensive manually annotated public data sets for mass spectrometry data are labor-intensive to produce and their quality strongly depends on the skill of the human expert. Some parts of the data may even be impossible to annotate due to high levels of noise or other ambiguities. Furthermore, manually annotated data is usually not available for all steps in a typical computational analysis pipeline. We thus developed the most comprehensive simulation software to date, which allows to generate multiple levels of ground truth and features a plethora of settings to reflect experimental conditions and instrument settings. The simulator is used to generate several distinct types of data. The data are subsequently employed to evaluate existing algorithms. Additionally, we employ simulation to determine the influence of instrument attributes and sample complexity on the ability of algorithms to recover information. The results give valuable hints on how to optimize experimental setups. Furthermore, this thesis introduces two quantitative approaches, namely a decharging algorithm based on integer linear programming and a new workflow for identification of differentially expressed proteins for a large in vitro study on toxic compounds. Decharging infers the uncharged mass of a peptide (or protein) by clustering all its charge variants. The latter occur frequently under certain experimental conditions. We employ simulation to show that decharging is robust against missing values even for high complexity data and that the algorithm outperforms other solutions in terms of mass accuracy and run time on real data. The last part of this thesis deals with a new state-of-the-art workflow for protein quantification based on isobaric tags for relative and absolute quantitation (iTRAQ). We devise a new approach to isotope correction, propose an experimental design, introduce new metrics of iTRAQ data quality, and confirm putative properties of iTRAQ data using a novel approach. All tools developed as part of this thesis are implemented in OpenMS, a C++ library for computational mass spectrometry.