Prediction of Genomic Methylation Status on CpG Islands Using DNA Sequence Features

-. In mammals, cytosines of most CpG dinucleotides in their genomes except gene promoters are subject to modification by methyl group (methylation). A number of genes in a mammal are regulated developmentalspecifically or tissue-specifically by the methylation. Mammalian DNA methylation contributes to regulation of gene expression, repression of parasitic sequences, inactivation of X chromosome in female, genomic imprinting, etc. Aberrant methylation results in a part of cancers and genetic diseases in human. Therefore it is required that methylation status on human genome is comprehensively revealed in each kind of cells. However, since comprehensive methylation analyses require a lot of times and large labor, methylation status on only a part of genomic regions is revealed in mammals. Because of this, machine learning using already known methylation data and prediction of methylation status on other genomic regions are important. Moreover, since sequence differences between DNA regions showing different methylation status also remain unclear, those differences should be also determined. Therefore we conducted machine learning by support vector machine using our previously reported methylation data, and predicted methylation status on DNA sequences using DNA sequence features. Furthermore we explored different sequence features among four types of methylation using random forest. Consequently high methylation prediction accuracies were observed between two different methylation status pairs. Moreover it was revealed that sequences containing CG, CT or CA were important for discrimination between them. Key-Words: CpG island, DNA methylation, Human chromosome 11, Human chromosome 21, Support vector machine, Random forest