PREDICTIVE ANALYSIS OF CONDITIONAL EPIGENETIC VARIABILITY
Ahmet Sinan Yavuz
Molecular Biology, Genetics and Bioengineering, PhD Dissertation, 2017
Prof. Dr. İsmail Çakmak (Thesis Advisor), Prof. Dr. Osman Uğur Sezerman, Prof. Dr. Yücel Saygın, Assoc. Prof. Devrim Gözüaçık, Asst. Prof. Özgür Asar
Date & Time: March 10th, 2017 – 09:00 AM
Place: FENS L061
Keywords: DNA methylation, cancer genomics, machine learning, random forests
DNA methylation is one of the most studied epigenetic mechanisms, as it functions in transcriptional regulation, cell differentiation, and genomic imprinting. Aberrant DNA methylation is shown to be one of the hallmarks of many complex diseases, such as Alzheimer's, Parkinson's, and various cancers. Specifically, global loss of methylation patterns and promoter-specific gain of methylation patterns were found to shape the epigenetic landscape for carcinogenesis and adaptation of tumours. Although disease-associated differential methylation regions are heavily studied, no attempt so far has been made to systematically examine the common characteristics of aberrant DNA methylation regions, and the mechanisms that lead the formation of these regions. In this dissertation, we have developed a random forest-based approach using a set of liver hepatocellular carcinoma patients to identify common characteristics of significant differential methylation events in a patient-specific manner. A variety of information, including features derived from region sequence, genomic selection and conservation, DNA shape, known regulatory features such as cis-regulatory elements, genome segmentation, histone modifications, and DNAse I hypersensitive sites, as well as, patient-specific mRNA and miRNA expression patterns, copy number variations, single nucleotide polymorphisms, and clinical variables were investigated. Lastly, protein-protein interaction networks were utilised to devise putative patient-specific and common mechanisms that may drive aberrant DNA methylation patterns.