Prediction and analysis of the methylation status of CpG islands in human genome
MetadataShow full item record
DNA methylation serves as a major epigenetic modification crucial to the normal organismal development and the onset and progression of complex diseases such as cancer. Computational predictions for DNA methylation profiling serve multiple purposes. First, accurate predictions can contribute valuable information for speeding up genome-wide DNA methylation profiling so that experimental resources can be focused on a few selected while computational procedures are applied to the bulk of the genome. Second, computational predictions can extract functional features and construct useful models of DNA methylation based on existing data, and can therefore be used as an initial step toward quantitative identification of critical factors or pathways controlling DNA methylation patterns. Third, computational prediction of DNA methylation can provide benchmark data to calibrate DNA methylation profiling equipment and to consolidate profiling results from different equipments or techniques. This thesis is written based on our study on the computational analysis of the DNA methylation patterns of the human genome. Particularly, we have established computational models (1) to predict the methylation patterns of the CpG islands in normal conditions, and (2) to detect the CpG islands that are unmethylated in normal conditions but aberrantly methylated in cancer conditions. When evaluated using the CD4 lymphocyte data of Human Epigenome Project (HEP) data set based on bisulfite sequencing, our computational models for predicting the methylation status of CpG islands in the normal conditions can achieve a high accuracy of 93-94%, specificity of 94%, and sensitivity of 92-93%. And, when evaluated using the aberrant methylation data from the MethCancerDB database for aberrantly methylated genes in cancer, our models for detecting the CpG islands that are unmethylated in normal conditions but aberrantly methylated in colon or prostate cancer can achieve an accuracy of 92-93%, specificity of 98-99%, and sensitivity of 92-93%.