Ziwei Huang: Using network-based and multi-omics approaches to study mechanisms in biological systems

Network-based approaches have been widely employed in biomedical research to depict system-level dependencies, model functional relationships and interpret complex biological processes at different omics. Importantly, the network structures underlying each cellular level play a key role in determining the functional properties of biological systems at that level. Integrative analysis of data across omics layers further facilitates the identification and interpretation of underlying mechanisms within biological systems. However, the absence of well-defined associations between metabolomics and other omics presents challenges in both the analysis of metabolomics data and its integration with other omics layers. To overcome this barrier, we integrated genome-scale metabolic models (GEMs), curated network models that collect all known metabolic information of a biological system, to develop hypeR-GEM, a methodology and associated R package that leverages the metabolic information encoded in GEMs to build reaction-based connections between metabolites and enzyme-coding genes. Additionally, we utilized Gaussian graphical models (GGMs), undirected, partial correlation-based networks that capture conditional dependencies between variables, to infer networks from data. However, the reliability of inferred network structures is often compromised by the limited sample sizes characteristic of high throughput “omics” data. To address this challenge, we propose RSCGGM-Resampling-based Consensus Gaussian Graphical Models-a framework and associated R package designed to enhance the reliability of GGMs inferred from high-dimensional data. This framework provides confidence intervals, as well as nominal and adjusted p-values for each edge, rather than a single point estimator, thereby enhancing the reliability of inferred network structures. We evaluated RSCGGM on simulated network structures representative of known biological properties, observing improved F1 scores that demonstrate its effectiveness. Furthermore, motivated by the inherent sign information, we incorporated structural balance theory (SBT) and a null model based on the maximum entropy framework that preserves both signed degrees and network topology, to assess structural balance at the level of triangle graphlets and thereby gain deeper insights into the underlying mechanisms of biological systems. Notably, we observed significantly balanced patterns in the networks inferred by RSCGGM from two independent breast cancer transcriptomics datasets. In our next steps, we will investigate the specific genes driving these patterns using metrics such as the graphlet degree vector and compare them with previously identified oncogenes to assess SBT's utility in analyzing biological networks further. Additionally, we plan to apply RSCGGM to proteomics and metabolomics data to study the mechanisms underlying healthy aging and longevity. Proposed Exam Time: Fall 2024 semeste