PDF, English - main document Restricted access: Repository staff only until 1 July 2024. Login+Download (42MB) | Terms of use

Abstract

Background: Machine learning approaches are becoming increasingly common in biological research, as these allow for a better understanding of the complex cell dynamics. Epigenetics encompasses processes able to modulate gene expression that do not depend on genomic sequence. Oftentimes, epigenetic alterations have been linked to disease. In this thesis, we applied several computational approaches to characterise the epigenetic landscape of diseased states caused by Human Immunodeficiency Virus infection and cancer in the brain.

Results: On the first part of this thesis, we applied non-negative matrix factorisation to build an epigenetic state map for the C20 microglial cell line and assessed the connection between integration and epigenetics in the context of HIV-1 infection. Through random forest models, we observed that genomic targets of HIV-1 integration are influenced by the initial epigenetic landscape and that infection leads to changes in the chromatin accessibility and TF binding. Furthermore, we found that regions often targeted by viral integration are associated to higher order chromatin structures, in particular topologically associated domains. On the second part of this thesis, we characterised CpG islands (CGI) of four glioblastoma subtypes and identified a new phenotype of CGI hypermethylation associated to RTK-II subtype, different from the one observed on the IDH subtype. We compared the CGI hypermethylation phenotypes associated to the IDH and RTK-II subtypes using random forests and use progenitor states to assess the tendency within each CGI to become hypermethylated. We observed that CGI most likely to become hypermethylated in cancer are marked already on undifferentiated cell states. Moreover, we observed that RTK-II CGI hypermethylation disturbs the astrogenic/neurogenic fate balance.

Conclusions: This thesis provides novel insights into the epigenetics of HIV-1 integration and CGI hypermethylation in glioblastoma. Through a genomic and epigenomic data-driven approach, we emphasise the importance of computational approaches like non-negative matrix factorisation, random forest, and bayesian networks into epigenetic research, as these provided an hollistic view of the global effects of viral integration and CGI hypermethylation in human cells.