Methods for discovering impartial profiles in metabolomics data

© 2016 Dr. Chalini Dilushika WijetungeMetabolomics is one of the main branches in systems biology, which aims at identifying and quantifying small molecules known as metabolites in biological samples. The metabolome of a living organism comprises of various metabolites such as organic acids, amino acids, vitamins and sugars that reveal plethora of valuable information on biological processes. Various Mass Spectrometry (MS) and Imaging Mass Spectrometry (IMS) based technologies are being utilized in the identification and quantification of metabolites in living systems. Although the technologies used for metabolomics data acquisition generate substantial amounts of data that are critical in solving various biological questions, the knowledge discovery and pattern recognition methods required to explore the resultant data are currently lagging behind, compared to the rapid technological advancements. Therefore, this research mainly focuses on developing robust unsupervised/semi-supervised data analysis methods for exploring metabolomics data. This thesis primarily attempts to address three research questions that are highly critical to solve several biological problems in the area of metabolomics, namely: 1) how to identify peaks and accurately estimate their parameters in MS data, 2) how to discover hidden significant metabolite distribution patterns from IMS data, and 3) how to identify metabolites suspected of contributing to an abnormality in living systems by comparing multiple IMS datasets. Peak detection is the most important step in MS metabolomics data analysis. Estimation of peak parameters such as peak summit location and peak area should be performed as accurately as possible in order to identify the underlying metabolites and to estimate their abundance precisely. The errors that occur in this process can vastly affect the final result thereby leading to wrong predictions. I developed an improved method for MS peak detection and parameter estimation, incorporating a new peak model named “mAPV”. The competency of this algorithm and the new peak model were validated using data generated from a Matrix Assisted Laser Desorption Ionization (MALDI) MS computer model and simulated data. The proposed algorithm is particularly beneficial for mass spectra having overlapping peak distributions and asymmetric peaks. Over the past decade, another data acquisition technique called MALDI-IMS has gained popularity in metabolomics, owing to its ability to provide information about the spatial distribution of metabolites, which was not available through other MS based analytical techniques. This additional information is highly beneficial in order to get a holistic view of the metabolome of any living organism. Most existing methods in IMS data analysis do not utilize the maximum benefit of the sheer volume of important spatial information generated by this technology. As an attempt to fill this gap, I developed an unsupervised data analysis pipeline incorporating a novel spatial distribution based peak picking method, for discovering hidden significant metabolite distribution patterns from MALDI-IMS metabolomics data. The proposed pipeline and the peak picking algorithm were evaluated using MALDI-IMS datasets of a plant leaf and a rat brain respectively. The ultimate goal of metabolomics studies is to discover metabolites suspected of con- tributing to an abnormality in living systems. Therefore, I proposed an improved method to identify causal agents of a plant/animal disease by using MALDI-IMS metabolomics data. The competency of the proposed algorithm was demonstrated using MALDI-IMS data of control and treated rat kidney samples. Apart from addressing the aforementioned research questions of utmost importance in metabolomics, I also did a case study on the performance of existing unsupervised learning methods in exploring MS metabolomics data, and proposed to use the Growing Self Organizing Maps (GSOM) algorithm in this regard. The competency of the GSOM algorithm in exploring MS data was validated by uncovering metabolites suspected of contributing to the drought tolerance mechanism of three south Australian wheat varieties