Please use this identifier to cite or link to this item:
A Metabolome-Microbiome Approach in the Study of Chronic Liver Disease
File under embargo until 2021-07-01
|Title:||A Metabolome-Microbiome Approach in the Study of Chronic Liver Disease|
|Contributors:||Jia, Wei (advisor)|
Molecular Biosciences and Bioengineering (department)
Chronic liver disease
show 4 moreMass spectrometry
|Publisher:||University of Hawai'i at Manoa|
|Abstract:||Chronic liver disease (CLD) is a continuing disease process with progressive destruction and regeneration of the liver parenchyma thus resulting in liver fibrosis, cirrhosis and even hepatocellular carcinoma (HCC). CLD is one of the leading causes of death in the United States which relates to various risk factors, e.g., hepatitis B/C virus infection, alcohol usage, high-fat diet. The liver fibrosis is reversible with early detection and proper intervention while even cirrhosis may be still regressive with control of underlying cause and treating with suitable medicine in time. Thus, the accurate diagnosis and staging of liver fibrosis for CLD patients are necessary. Currently, the gold standard for liver fibrosis diagnosis and staging is the liver biopsy which associates with various clinical limitations as an invasive approach. Thus, non-invasive measurements, e.g., blood-based test, provide a more convenient and practical method for continuing surveillance of CLD. In Chapter 1, we depict the research background of CLD and introduce some relevant topics that included in our study.|
Metabolomics has long been used for the diagnosis and prognosis of human diseases, which can provide a less-invasive approach to overcome certain limitations of the gold standard, facilitate medical decision making, and enhance long-term clinical surveillance of CLD. Towards this goal, a large-scale metabolomics study in CLD is needed to find a panel of robust biomarkers for the diagnosis and staging of hepatic fibrosis with strong statistical power. However, a large-scale metabolomics study poses various computational challenges to current metabolomics analysis pipeline. One computational challenge is the computing speed, considering the polynomial increase in computing time depending on the number of input files. Another challenge is the signal intensity drift on a long-running operation of the analytical platform that introduces unwanted variations to the metabolic signals and influences the real biological signals. In Chapter 2, we specifically discuss how we solved these challenges to facilitate the large-scale metabolomics study. We first redesigned a comprehensive parallel computing framework based on our original metabolomics data analysis framework. We observed significant decrement in computing time regarding raw data processing and identification. We then implemented a non-linear quality-control (QC) correction approach using locally estimated scatterplot smoothing (LOESS) regression based on a well-recognized metabolomics protocol and found the LOESS QC correction did reduce the signal drifts and outperformed the widely-applied block QC correction.
Another challenge widely occurred in mass-spectrometry (MS)-based metabolomics study is the existence of missing values, which impairs the statistical power and biases the results in a great deal. In Chapter 3, we systematically compared the existing missing value imputation approaches for different missing types and provided a comprehensive missing value processing approach with a free-accessible web-tool, i.e., MetImp, for other researchers. In addition, we developed a Markov chain Monte Carlo-based algorithm, GSimp, for the left-censored missing value imputation and outperformed all present imputation approaches in the situation of missing not at random. This algorithm was also included in our MetImp web-tool.
With the above improvements in metabolomics data processing and pretreatment, in Chapter 4, we report on the results of a large-scale clinical metabolomics study on hepatitis B virus (HBV)-related CLD to discover potential metabolic biomarkers for the diagnosis and staging of the hepatic disease. We employed several machine learning approaches to select predictive markers from quantitative metabolomics data sets. We then applied an ensemble learning model, random forest, to construct predictive models for the diagnosis of CLD, differentiation of fibrosis and cirrhosis, and detection of early-stage fibrosis. The metabolic marker-based panel tandem with cutting-edge machine learning algorithm demonstrated the best classification performances compared to other existing blood-test based clinical indices. The metabolites-based predictive models were then validated in two external independent validation data sets. This study confirmed that metabolic markers with machine learning model could serve as a powerful approach for the non-invasive diagnosis and staging of hepatic disease.
Moreover, we were curious whether we could augment existing blood-based clinical scoring system by applying proper machine learning algorithms. In Chapter 5, on an HBV-related CLD dataset, we describe the training of an ensemble learning model, gradient boosting that outperformed other machine learning models as well as the original clinical index, FIB-4. The model prediction was then validated in additional HBV and hepatitis C virus (HCV)-CLD datasets from previous published studies. This study further confirmed the strength of using machine learning algorithms as a powerful prediction tool for hepatic disease.
In addition to metabolic changes, gut-microbiota has been shown to play an essential role in the development of the chronic hepatic disease which might additionally serve as potential non-invasive biomarkers for the CLD. Additionally, the disturbance of metabolome-microbiome association networks might reflect the fluctuations of the liver-gut axis regarding CLD. In Chapter 6, we describe both MS-based metabolomics and amplicon sequencing-based gut-microbiome studies on the streptozotocin-high fat diet-induced nonalcoholic steatohepatitis (NASH) and hepatocellular carcinoma (HCC) mice as well as matched control mice. Group differential bacteria were identified and co-occurrence networks were altered according to the hepatic disease. The microbiome-metabolome association analysis revealed a significant disturbance of the microbiota-bile acids cross-talk along with NASH-HCC. Although based on a mice model, this study provided relevant insights that how gut-microbiota and host-metabolome could change according to the progression of CLD.
In the last chapter, Chapter 7, we summarize all of our current studies in the CLD. We then discuss the potential limitations and challenges associated with our project and propose possible future directions for the followed-up studies.
|Description:||Ph.D. Thesis. Ph.D. Thesis. University of Hawaiʻi at Mānoa 2019|
|Rights:||All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.|
|Appears in Collections:||
Ph.D. - Molecular Biosciences and Bioengineering|
Please email firstname.lastname@example.org if you need this content in ADA-compliant format.