Tao Huan

This thesis delves into metabolomics, the study of small molecules—known as metabolites—present within a biological system. These molecules span both polar classes, such as amino acids, sugars, and nucleotides, and non-polar classes, such as fatty acids and other lipid species, which are often collectively referred to as lipids. The study of this latter group is often treated as a distinct domain called lipidomics, which focuses specifically on the comprehensive analysis of lipids. Accordingly, the term “metabolomics” may be used either as a broad umbrella that includes both polar and non-polar molecules, or a narrower term that refers to the study of polar molecules only. This work adopts the latter definition and focuses on integrating metabolomics with lipidomics to enable a more comprehensive investigation of the biological systems of interest. The first project focuses specifically on lipidomics, examining the lipid composition of extracellular vesicles secreted by wild-type and drug-resistant Leishmania parasites. By annotating and comparing lipid species present in these vesicles, this study provides detailed insight into lipid remodeling associated with drug resistance. Extracellular vesicles play a critical role in parasite–host interactions, and alterations in their lipid composition may influence membrane properties, signaling processes, and resistance mechanisms. This lipid-centric approach enables high-resolution chemical profiling of resistance-associated changes and contributes to the identification of lipid signatures that may serve as biomarkers or therapeutic targets in leishmaniasis.The second project expands beyond a single molecular class and addresses a key methodological challenge relevant to both metabolomics and lipidomics: the comparison of single-phase and dual-phase extraction strategies. While dual-phase extraction enables the simultaneous recovery of polar metabolites and lipids from a single sample, it may result in reduced analyte concentrations compared to single-phase methods. This study systematically investigates three mechanistically distinct factors hypothesized to drive these differences—pipetting effect, partitioning effect, and matrix effect. By isolating and evaluating each factor, this work provides a framework for objectively assessing extraction performance and informs methodological decision-making in integrative small-molecule analyses.Overall, this thesis advances the field by combining a lipidomics-focused biological application with a broader methodological evaluation relevant to integrated metabolomics–lipidomics workflows.

View record

Metabolomics, the study of small molecules within biological systems, offers valuable insights into biochemical processes. This thesis addresses two challenges in mass spectrometry-based metabolomics: processing complex breathomics data and enhancing the structural annotation of metabolites using deep learning.In the first study, I introduce BreathXplorer, an open-source Python package designed to process real-time exhaled breath data from secondary electrospray ionization high-resolution mass spectrometry (SESI-HRMS). BreathXplorer tackles the challenge of non-Gaussian metabolic signal shapes by employing topological algorithms or Gaussian mixture models (GMM) to identify exhalation intervals and density-based spatial clustering (DBSCAN) to cluster m/z values. It accurately determines the start and end points of exhalation, ensuring precise quantitative measurements. In a proof-of-concept study on exercise breathomics, BreathXplorer identified exercise-responsive metabolites, showing its potential in real-time metabolomics research.In the second study, I explore deep learning as a solution for compound annotation in tandem mass spectrometry (MS/MS). This offers a predictive strategy where traditional library search tools are limited due to small spectral libraries. Generative models have become a fundamental approach for various tasks; however, their application to MS/MS structural annotation is less developed and often underperforms compared to traditional fingerprint-based counterparts. In this work, I investigate the potential of generative models by building and comparing transformer-based fingerprint models and generative models. This comparison helps to understand their strengths and limitations in annotating chemical structures. Training and testing on 616,594 unique structures, I identified three key limitations of direct structure generation with generative models: (1) error accumulation, (2) generation of invalid compounds, and (3) generation of unrecorded compounds. I then propose a solution using the generative model as a ranker. My results demonstrate that generative-based ranking outperforms fingerprint-based systems. Furthermore, I observed a positive correlation between high generation scores and the correctness of predicted structures. My analysis shows that generative models can leverage richer structural information during training, leading to improved accuracy in end-to-end chemical structure identification.

View record

This thesis delves into the development and application of metabolomics, a discipline focused on the comprehensive study of metabolites within biological systems. The research is segmented into two interconnected parts: methodology development for metabolomics data processing and its subsequent application in plant stress physiology. The first project tackles the challenge of computational variation in untargeted metabolomics, which arises due to the incapability of data processing for complex LC-MS data. An in-depth exploration led to the identification of sources and causes of computational variation, followed by the development of novel methodologies to mitigate these challenges. These methodologies, including data processing parameter optimization and a machine learning program, successfully reduced computational variation, thereby enhancing the quantitative precision of untargeted metabolomics. The second segment of the thesis applies these methodologies to study the salinity stress response in Alfalfa (Medicago sativa L.). Comprehensive analysis of the plant's metabolic alterations, coupled with transcriptomics data, revealed significant pathways and mechanisms of salinity response. The integration of multi-omics data provided a deeper understanding of the complex interplay between genes and metabolites. The research advances the field of metabolomics, providing improved data processing methodology and valuable insights into plant stress physiology. Future work may expand these findings towards personalized medicine, disease diagnosis, and precision treatment.

View record

Metabolomics and exposomics are rapidly expanding fields that aim to understand the intricate relationship between environmental exposures and human health. Hair is an underexplored matrix for studying metabolomics and exposomics, serving as a record of chemicals deposited on its surface. However, the lack of a comprehensive database and compound annotation pipeline has hindered the use of hair in untargeted studies. To address these challenges, a comprehensive database of hair metabolomes and exposomes, HairDB, was introduced in Chapter 2. The database systematically compiles all reported hair chemicals through text-mining and manually incorporated chemicals with a high likelihood of being deposited on the surface of hair. HairDB contains 4191 unique chemicals from 9214 articles, 172 of which were further categorized as biomarkers. The user-friendly web interface of HairDB will facilitate the widespread adoption of hair as a matrix for research. Chapter 3 presents a novel bioinformatic pipeline for global-scale hair exposome studies, including optimized feature acquisition, feature quality assurance, intensity correction, top-down annotation, and molecular formula prediction using BUDDY. Out of the 26119 features detected, 316 were annotated using NIST20 and MS-DIAL spectral libraries. BUDDY predicted molecular formulas for 4533 unidentified metabolic features, with 4270 being successfully predicted. Global optimization of chemical annotation was applied to detect 1755 potential transformations, with 579 being between identified metabolic features and those with predicted molecular formulas through known chemical reactions. Using HairDB, 43 unique metabolites were found with corresponding literature. The predicted molecular formulas were also used to search HairDB, resulting in 275 hits. The development of HairDB and the annotation pipeline for hair metabolome offer valuable resources for researchers in the fields of metabolomics and exposomics.

View record

As the most recently emerged “omics”, metabolomics grabbed attention in human health studies by measuring thousands of small-molecule metabolites in a wide range of biological samples. As the downstream products in the biological pathway, metabolites are regarded as the closest link to the phenotypes. Small stimuli in the human body will cause relatively huge changes in the level of metabolites. Liquid chromatography-mass spectrometry (LC-MS) is the mainstay in metabolomics research due to its high throughput, sensitivity, and reliable analysis of metabolites. Nevertheless, two of the main challenges in LC-MS based metabolomics are 1) how to apply metabolomics in studying human health and 2) apart from commonly used biological samples, including serum, plasma, and urine, how to develop a methodology of new biological samples that can be adapted to specific human health research. To address those challenges, in Chapter 2, I integrated metabolomics with metagenomics to examine human gut health. 13-species metagenomic signature was selected by random forest machine learning and achieved high diagnostic accuracy in differentiating hepatic decompensation in NAFLD-related cirrhosis. The signature was cross-validated by metabolomics. 32 metabolites and 15 metabolites from serum and feces, respectively, were found to be significantly linked to 13-discriminatory species, suggesting that the identified discriminatory species may play important roles in the progression from compensated to decompensated cirrhosis. This multi-omics study yields new avenues for identifying novel targets for therapy and microbial biomarkers of hepatic decompensation, a worldwide human disease. In Chapter 3, I integrated plasma metabolomics and proteomics to examine the health conditions of highly trained females and males following acute, severe-intensity exercise. Metabolomic and proteomic homeostasis were substantially perturbed. Through statistical analysis, some metabolites and proteins were found to be closely linked to high-intensity exercise. This multi-omics study was a powerful tool to study molecular responses to acute exercise and provided a new insight to exercise-bolstered human health. In Chapter 4, I developed a new methodology to track skin secretion. Our high-performance workflow was readily applied to a wide range of skin metabolomics research to gain a better understanding of the molecular signatures on skin that link to human health and disease.

View record

For this work, two areas of metabolomics were investigated relating to the fundamentals of the field and application to different experiments. The first chapter was an assessment and comparison of the MSMS spectra generated from different acquisition modes. The chosen acquisition modes were data-dependant acquisition (DDA), data-independent acquisition (DIA), and enhanced insource-fragmentation (eISF) at a range of collision energies. The data was obtained by performing untargeted metabolomics on a urine sample and a standard mixture solution through a LC-MS platform while also covering multiple ionization modes. The spectra from the three modes were compared against each other through several factors that relate to the various ways MSMS spectra are used in a metabolomics workflow. These comparisons involved investigating the spectral purity, quality of reference matching results, structural similarity, and de novo annotation performance. It was found that DDA performed the best with eISF and DIA following. It was seen that eISF performed on-par or slightly better than DIA at higher collision energies. This indicates that the collision energy used will have a notable impact on the performance of the mode. The second chapter involves the metabolomics of pancreatic cell samples. The purpose of this was to determine the metabolic profile between the control and treated groups. The control was regular cancer cells from the MiaPaCa2 cell line while the treated groups had specific genes knocked out. The investigation was performed to gain insight into which metabolic pathways the knocked-out genes were involved in. Using a LC-MS platform it was found that 12 metabolites showed significant intensity differences between the groups. A literature review of these compounds highlighted possible metabolomic pathways affected such as polyamine metabolism. The last chapter focuses on a lipidomics experiment that was performed on the bacteria Thermotoga maritima to investigate the lipid content of the bacterial membranes. The samples relating to each fraction were run through the same LC-MS platform as above. It was seen that there were three significantly different lipids apart from the fatty acid, phosphatidylethanolamine, and phosphatidylinositol lipid classes. These classes have all been shown to be involved in membrane stability and transport.

View record