Paper
A Predictive Model for Genomic Methylation Targets in Humans
- Authors:
- Andigoni Malousi; Ioanna Chouvarda; Sofia Kouidou; Nicos Maglaveras
- Abstract
- DNA methylation is a tissue-specific, sequence-dependent and dynamic process, regulated by complex developmental mechanisms. Recent genome-wide studies provide compelling evidence for long-range sequence-dependence of this process, thus necessitating the computational identification of cytosine methylation and its deviations at individual CpG sites, processes which might reveal the activation of various epigenetic regulatory mechanisms. The present study deals with the prediction of methylation at single CpG sites located both in CpG islands and non-CpG islands, relying solely on the intrinsic information of the 401nt sequences centred at the candidate methylated CpG cytosines. The proposed method was evaluated on human methylation data available by the Human Epigenome Project. The proposed Sequential Minimization Optimization classifier, when trained with the most informative features, gives highly accurate identification of the hypermethylated and hypomethylated sequences (Sn=85.35%, Sp=90.72%). Post-classification analysis of the correctly and mistakenly classified methylation targets revealed that the classification accuracy varies depending on the co-localization of the CpG sites with genomic features involved in gene expression, such as CpG islands and exonic/intronic sequences. This observation reinforces the importance of incorporating site-specific features in the predictive model and the development of site-specific classifiers. Datasets and source code are available upon request.
- Keywords
- Predictive Model; DNA Methylation; Local Context; Sequential Minimization Optimization; Post-Classification Analysis
- StartPage
- 7
- EndPage
- 19
- Doi