Boston University
Network-based approaches to mechanistic discovery from multi-omics data
Abstract
dc:description.abstractHigh-throughput and multi-omics technologies generate rich, complementary data that enable systems-level investigation of molecular mechanisms in human biology and disease. Yet integrating heterogeneous omics layers and analyzing these high-dimensional datasets remain challenging. Network-based methods offer a natural framework to model system-level dependencies and fuse multiple omics modalities, but their effective application requires methodological advances to address integration, validation, and inferential limitations. In particular, statistical power for network inference is often constrained by modest sample sizes, and principled benchmarking frameworks for evaluating network methods are scarce. This dissertation develops methodological and computational frameworks to address these challenges. First, I introduce \textit{hypeR-GEM}, a methodology and associated R package that adapt gene set enrichment analysis for metabolomics by leveraging genome-scale metabolic models to infer reaction-based relationships between metabolites and enzyme-coding genes, thereby facilitating metabolomics integration with other omics layers across diverse biological contexts. Second, I present \textit{RSNet}, an open-source R package implementing a resampling-based framework for robust network inference and downstream analysis, supporting Gaussian graphical models and conditional Gaussian Bayesian networks for mixed data types. Third, I develop a principled benchmarking framework for evaluating structure learning in Gaussian graphical models. Drawing on random matrix theory and high-dimensional statistics, I propose a Wigner-based precision-matrix generator that enables analytical characterization of network-detectability properties, ensuring observed performance differences reflect methodological properties rather than artifacts of the data-generating process. Together, these contributions advance both the methodological foundations and practical applications of network-based analysis for high-dimensional multi-omics datasets, improving the robustness, interpretability, and reliability of computational approaches in biomedical research.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Physics
- Grantor
- Boston University
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Huang, Ziwei
- Advisor dc:contributor.advisor
-
- Monti, Stefano
- ORCID dc:identifier.orcid
Subjects
dc:subject × 7Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/2144/53738