Machine Learning-Guided, Biomarker-Enabled Disease Progression Modeling

• Main Author: McComb, Mason Charles
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01.01.2020
• Subjects: Artificial intelligence; Bioinformatics; Pharmaceutical sciences
• ISBN: 9798582508441

This dissertation investigates strategies for delineating the mechanisms and dynamics of disease progression that utilize innovative computational, statistical and mathematical strategies in conjunction with clinical and biomarker data. Structurally, this dissertation has two distinct but complementary foci:1. Development, proof-of-concept and integration of cutting-edge machine learning-guided strategies for disease progression modeling in pharmacometrics (PMX).2. Mechanistic evaluation of novel biomarkers for disease progression modeling in multiple sclerosis (MS). The first focus of the dissertation is contained within Chapter 1, 2 and 3. Chapter 1 reviews the state of the art of machine learning (ML) research for PMX and pharmaceutical sciences applications. The review indicated that ML has a promising utility for leveraging “big data” in PMX modeling and can be used as a complementary technique to aid model structure development and covariate identification during drug development.The utility of ML techniques for covariate modeling in population PMX was further assessed using a framework that coupled random forest regression with Bayesian networks. This integrated approach identified subject-specific factors associated with cholesterol dynamics. The results provided a roadmap for enhancing covariate identification for PMX modeling using ML, and for modeling drug outcomes in diverse populations. The integrated ML-guided framework for harnessing “big data” was extended to modeling the natural history of disease-relevant biomarkers by including a stochastic modeling component. Discrete-time vector autoregression models and multivariate stochastic differential equations were investigated in the stochastic modeling component. The stochastic models incorporated regulatory inter-dependencies among the biomarkers and an aging process. The integrated “big data”, ML and stochastic modeling approach was capable of describing the dynamics of three sets of disease-relevant biomarkers involved in metabolism and cardiovascular disease that had complex dynamics across the lifespan. The second focus of the dissertation, described in Chapters 4- 7, is a mechanistic investigation and statistical modeling of the multifaceted aspects of neurodegeneration and pathophysiological processes and in multiple sclerosis (MS). The research is specifically directed at leveraging serum neurofilaments (sNfL) levels, which are increased during neuroaxonal injury, to obtain a better understanding of the mechanisms of MS neurodegeneration and disease progression. In Chapters 4 and 5, sNfL measurements were used to understand the role of the cholesterol pathway in disease progression in earliest stages of MS following the first demyelinating event. Several magnetic resonance imaging measures, cerebrospinal fluid (CSF) biomarkers of blood-brain barrier (BBB) integrity and central nervous system (CNS) inflammation were used to evaluate the potential of sNfL as a surrogate disease progression biomarker. The association of sNfL levels were associated with immune cell-extravasation and brain lesion activity on MRI in MS patients at the first demyelinating event. In the next chapter, the neuroprotective pathophysiological mechanisms of cholesterol pathway biomarkers on BBB integrity, CNS inflammation and neuroaxonal injury, as assessed by CSF and serum NfL measurements, were investigated. The cholesterol biomarkers apolipoprotein A-II (ApoA-II) and ApoA-I, which are characteristic proteins present on high-density lipoprotein cholesterol (HDL-C) particles, were negatively associated with CSF NfL and sNfL levels, demonstrating a neuroprotective role for ApoA-II and ApoA-I in early MS. Chapters 6 and 7 focus on the role of sNfL in MS patients with longer disease duration. The role of longitudinal changes in lipid biomarkers and sNfL levels over the 5-year follow-up duration on MS neurodegeneration and disease progression was investigated in relapsing-remitting and progressive MS patient groups. Upon adjusting for baseline sNfL, percent changes in HDL-C were associated with decreased gray matter atrophy and cortical atrophy. In Chapter 7, the associations of sNfL levels with non-enzymatic and enzymatically produced cholesterol oxidation products were assessed. The oxysterols, 7-ketocholesterol and 7β hydroxycholesterol, which are produced by reactive oxygen species (ROS)-mediated cholesterol oxidation, were positively associated with sNfL levels. Thus, oxysterols produced by ROS-mediated cholesterol oxidation, are differentially associated with neuroaxonal injury compared to the enzymatically generated oxysterols and that these associations are not explained by low density lipoprotein-cholesterol or HDL-C. levels.The concluding chapter of the dissertation bridges the themes of the two foci by summarizing the potential impact of the research on disease progression modeling and suggesting priorities for future research.