A Study of DNA Methylation Through the Lens of Machine Learning
Abstract
DNA methylation is an epigenetic modification that influences a wide range of critical cellular processes, such as gene expression, genome stability, transposon silencing and gene imprinting. In humans, the dysregulation of DNA methylation Is associated with a variety of diseases, including cancer and neurological disorders.
This thesis investigates several computational problems related to the analysis of DNA methylation.
In the first chapter, we introduce a method for the prediction of DNA methylation from the underlying DNA sequence, across different cytosine contexts. By leveraging information from co-located genomic functional elements, our deep-learning method outperforms state-of-the-art approaches.
In the second chapter, we study DNA methylation patterns in two malaria parasites, identifying potential avenues for drug discovery against this debilitating disease.
In the third chapter, we propose a method that can predict the activity of enzymes responsible for DNA methylation and demethylation in human and mouse embryonic stem cells directly from the sequence context. Our study elucidate new sequence motifs associated with the prediction of differentially methylated cytosines, casting a new light on the understanding of DNA methylation dynamics in stem cells.
In the last two chapters, we study the role of DNA methylation in cancer. We introduce a method for accurate thyroid cancer diagnosis and investigate the phenomenon of age acceleration induced by cancer in various tissue types.