Medizinische Universität Graz Austria/Österreich - Forschungsportal - Medical University of Graz
Gewählte Publikation:
Libiseller, G.
IPO: A Tool for automated Optimization of XCMS Parameters
Doktoratsstudium der Medizinischen Wissenschaft; Humanmedizin; [ Dissertation ] ; 2015. pp.
[OPEN ACCESS]
FullText
- Autor*innen der Med Uni Graz:
- Betreuer*innen:
- Pieber Thomas
- Sinner Frank Michael
- Sourij Harald
- Altmetrics:
- Abstract:
- Untargeted metabolomics aims to detect as many low-molecular-weight substances in a biological sample as possible and thereby generates a fingerprint of the sample. The fingerprint enables conclusions about the metabolic processes to be drawn and subsequently to determine the impact of metabolites on specific diseases and also the influence diseases have on metabolites. To detect these metabolites modern, high-resolution mass spectrometers are used. When they are coupled to gas or liquid chromatography a large amount of data is generated. Due to the enormous amount of data manual processing is not feasible. Therefore, programs have been developed to automatically filter noise, identify and correct signals in the data and merge the signals into meaningful metabolic features. In order to be as flexibly as possible and adaptable to various types of data these software products provide different parameters. The right choice of parameter settings is not trivial and has great influence on the reliability of the resulting data. So far, only a few theoretical approaches for automated optimization of these parameter settings have been published. A program that automatically performs parameter optimization and provides the best possible settings has not yet been developed. Therefore the aim of this thesis was to implement a software tool that automatically performs an optimization of parameter settings for any type of data. The parameters of the prevailing open-source software XCMS were to be optimized. To gain best possible parameters of peak detection in a first step a new score was developed. This score was based on the identification of signals derived from natural, stable carbon isotopes. These signals and the corresponding signals without carbon isotope were defined as reliable and led to the calculation of a score. In a next step, an additional score was defined which aimed to decrease the mean, relative retention time shifts within metabolic features while also increasing the number of reliable metabolic features. The various parameter settings were chosen and evaluated using a design of experiment approach. By using efficient design of experiments and splitting the optimization approach into two separate steps the number of necessary experiments was reduced to a minimum which consequently reduces the time needed for the optimization. The developed software has been tested on four different datasets originating from different types of samples which were measured with different devices and methods. All datasets showed a significant increase in the scores and thus an increased reliability of the data was observed. The developed software tool was published in the journal BMC Bioinformatics and is freely available from www.github.com/glibiseller/IPO.