Understanding gene regulation through the analysis of omics data

The interactions between chromatin, transcription factors, and genes form intricate regulatory circuits, which can be modeled as gene regulatory networks (GRNs). Historically, GRNs have been inferred from bulk profiling omics data, as well as from literature sources. The emergence of single-cell multi-omics technologies has driven the creation of many novel computational methods that integrate genomic, transcriptomic, and chromatin accessibility data, allowing in principle to infer GRNs at better resolution. In the first chapter of this thesis, I describe the classic and new approaches to measure and model gene regulation through GRNs and their downstream applications. In the second chapter, I describe the development of decoupler, a computationally scalable framework for the inference of TF activities from omics data through the pairing of enrichment analysis with GRNs. There I also compare several enrichment methods and conclude that simple linear models outperform classic enrichment methods. Then, I showcase how decoupler together with transcription factor activity inference can be used to discover new biological insights in human diseases. In the third and last chapter, I showcase the design and implementation of Gene Regulatory nETwork Analysis (GRETA), a comprehensive cross-method benchmark of multimodal GRN inference, and compare their performance relative to several baselines. There I show that although the obtained GRNs have predictive properties and can moderately recover known biology, they do not exhibit causal properties, contrary to what is always assumed of them. Additionally, I show how they perform on par, or worse than literature-derived GRNs or GRNs inferred only from transcriptomics, suggesting that inferring de-novo regulatory programs might be an overly complex problem and that the incorporation of biological knowledge could aid in GRN inference.