Search

Untargeted metabolomics by mass spectrometry in the form of mass over charge and intensity of ions, provides insight into the metabolic activity in a sample and is therefore essential to understand regulation and expression at the protein and transcription level. Problematically, it is often challenging to analyze untargeted metabolomics data as many m/z values are detected per sample and it is difficult to identify what compound they represent. We aimed to facilitate the process of finding m/z biomarkers through statistical analysis, machine learning and searching for their putative identities. To address this challenge, we developed MetaboShiny, a novel R and RShiny based metabolomics data analysis package. MetaboShiny features bi/multivariate and temporal statistics, an extensive machine learning module, interactive plotting and result exploration, and compound identification through a variety of chemical databases. As a result, MetaboShiny enables rapid and rigorous analysis of untargeted metabolomics data as well as target identification at unprecedented scale. To demonstrate its efficacy and ease-of-use, we apply MetaboShiny to a publicly accessible metabolomics dataset generated from the urine of smokers and non-smokers. Replication of the main results of the original publication, which includes importing, normalization and several statistical analyses, is achieved within minutes. Moreover, MetaboShiny enables deeper exploration of the data thereby revealing novel putative biomarkers and hypotheses. For instance, by using MetaboShiny’s subsetting feature, iodine is found to be significantly increased in non-smoking lung cancer patients. Furthermore, by allowing for custom adducts, MetaboShiny reveals a putative identification for an m/z value which could not be identified by the original authors. This validates MetaboShiny as a flexible and customizable data analysis package that greatly enhances metabolomics biomarker discovery.

The human gut microbiota plays a large role in the metabolism of our diet. These microorganisms can break down indigestible materials such as polysaccharides and convert them into metabolites that the human body can take up and utilize (e.g., vitamins, essential amino acids, and short-chain fatty acids). Disbalances in the gut microbiome have been associated with several diseases, including diabetes and obesity. However, little is known about the detailed metabolic crosstalk that occurs between individual organisms within the microbiome and between the microbiome and the human intestinal cells. Because of the complexity of the intestinal ecosystem, these interactions are difficult to determine using existing experimental methods. Constraint-based reconstruction and analysis (COBRA) can help identify the possible metabolic mechanisms at play in the human gut. By combining mathematical, computational, and experimental methods, we can generate hypotheses and design targeted experiments to elucidate the metabolic mechanisms in the gut microbiome. In this thesis, I first applied comparative genomics to analyze the biosynthesis pathways of eight B-vitamins in hundreds of human gut microbial species. The results suggested that many gut microbes do not synthesize any B-vitamins, that is, they depend on the host’s diet and neighboring bacteria for these essential nutrients. Second, I developed a semi-automatic reconstruction refinement pipeline that quickly generates biologically relevant genome-scale metabolic reconstructions (GENREs) of human gut microbes based on automatically generated metabolic reconstructions, comparative genomics data, and data extracted from biochemical experiments on the relevant organisms. The pipeline generated metabolically diverse reconstructions that maintain high accuracy with known biochemical data. Finally, the refined GENREs were combined with metagenomic data from individual stool samples to build personalized human gut microbiome metabolic reconstructions. The resulting large-scale microbiome models were both taxonomically and functionally diverse. The work presented in this thesis has enabled the generation of biologically relevant human gut microbiome metabolic reconstructions. Metabolic models resulting from such reconstructions can be applied to study metabolism within the human gut microbiome and between the gut microbiome and the human host. Additionally, they can be used to study the effects of different dietary components on the metabolic exchanges in the gut microbiome and the metabolic differences between healthy and diseased microbiomes.