Network-based approaches to understand transcriptional regulation in disease

Gene regulatory networks (GRNs) comprise the collection of regulatory relationships between molecular regulators such as transcription factors and their target genes. These networks are important for controlling fundamental biological processes, including development metabolism, and response to environmental stimuli. The inference and analysis of GRNs can benefit both basic biological research and practical applications, including precision medicine and drug repurposing. Given the complexity and context specificity of regulatory relationships across different disease conditions, new computational methods for analyzing high-dimensional omics data are warranted. This thesis presents two new approaches for the inference and analysis of GRNs applied to gene expression data. Diseases can induce perturbations in normal gene co-expression patterns. Thus, detecting differentially co-expressed or rewired edges between disease and healthy biological states can be useful for investigating the link between specific molecular alterations and disease phenotypes. The first study presents BoostDiff (Boosted Differential trees), a tree-based ensemble algorithm for inferring differential GRNs from two conditions. BoostDiff modifies standard regression trees by building an adaptively boosted (AdaBoost) ensemble of differential trees with respect to a target condition. Each differential tree serves as the base learner and is built using a new splitting criterion called differential variance improvement. The performance of BoostDiff is demonstrated using both simulated and real-world transcriptomics datasets. BoostDiff identifies context-specific networks that are enriched with genes of known disease-relevant pathways and can complement standard differential expression analyses. The second study explores the utility of combining active module identification on protein-protein interaction networks and the prize-collecting Steiner tree algorithm on GRNs. The proposed workflow, named Transcriptional Regulator Identification using Prize-collecting Steiner trees (TRIPS), is designed to identify candidate regulators influencing disease using results from differential expression analysis and prior knowledge from biomedical databases. Systematic analyses using multiple disease-associated datasets demonstrate that TRIPS can recover relevant regulators with high precision. Furthermore, our findings reinforce the importance of performing network perturbation analyses to characterize the reliability of network-based tools. In particular, the prize-collecting Steiner tree approach implemented in TRIPS was found to better utilize the knowledge encoded in the structures of molecular networks. TRIPS can be used as a post-processing step in analyzing gene expression data to provide network-based context for prioritizing transcriptional regulators in downstream experiments. Collectively, these two studies expand the computational toolbox for deriving biologically meaningful insights from omics data and can find practical application in the field of systems medicine.