Investigating the chromatin dynamics that underlie CRISPRi-validated enhancer-promoter interactions
Repository URI
Repository DOI
Change log
Authors
Abstract
Discovered in the 1980s, enhancers have grown to become important cis-regulatory elements responsible for both temporal and tissue-specific gene regulation coordinating cell differentiation processes, pathological responses and disease progression, and cellular functions in normal physiological states. However, due to their complex nature and function, research into enhancer activity remains highly active, producing novel insights based on multiple types of data and experimental assays. Firstly, advances in Next Generation Sequencing technologies in the 2000s allowed extensive enhancer characterisations through mass detection of genome wide epigenetic marks, chromatin conformation, chromatin accessibility, and expression data. Nevertheless, the process of validating enhancer targets remained costly until recent developments of CRISPRi methods enabled large scale perturbation experiments for identifying the target genes of enhancers. However, CRISPRi methods produced results that were not always consistent with the previous understanding of enhancer-promoter interactions (EPIs), raising questions regarding the mechanism by which EPIs are produced.
Then, this thesis aims to investigate the ways by which chromatin dynamics enable the function of CRISRPi-validated EPIs, in order to explain the discrepancies observed. Firstly, using experimental data from mouse cells, I consider EPIs based on chromatin folding data while highlighting the limitations of current frameworks. Then, I consider more complex representations of biological features to propose predictive models for improving EPI detection. Next, to investigate the chromatin dynamics that underlie CRISPRi-validated EPIs, I use unsupervised machine learning to identify distinct biological classes of EPIs that show mechanistic differences. With respect to these classes, I propose that they are underlied by different chromatin structures, and their functions may be related to enabling cancer progression. Thus, the former insight motivates the development of a graph-based deep learning model that can model complex chromatin structures and transcription factor binding data in order to predict EPIs and interpret their structures. Similarly, the second insight motivates the design of a CRISPRi experiment that can be used to validate the classification system in a second independent cell type. Then, I conclude that the relationship between the functions of CRISPRi-validated EPIs and chromatin dynamics reveals novel complexities that can be modelled computationally to describe important processes in gene regulation.