School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China.
Information and Computer Engineering College, Northeast Forestry University, Harbin, China.
BMC Genomics. 2020 Mar 5;21(Suppl 1):672. doi: 10.1186/s12864-019-6019-0.
Methylation of cytosine bases in DNA is a critical epigenetic mark in many eukaryotes and has also been implicated in the development and progression of normal and diseased cells. Therefore, profiling DNA methylation across the genome is vital to understanding the effects of epigenetic. In recent years the Illumina HumanMethylation450 (HM450K) and MethylationEPIC (EPIC) BeadChip have been widely used to profile DNA methylation in human samples. The methods to predict the methylation states of DNA regions based on microarray methylation datasets are critical to enable genome-wide analyses.
We report a computational approach based on the two layers two-state hidden Markov model (HMM) to identify methylation states of single CpG site and DNA regions in HM450K and EPIC BeadChip. Using this mothed, all CpGs detected by HM450K and EPIC in H1-hESC and GM12878 cell lines are identified as un-methylated, middle-methylated and full-methylated states. A large number of DNA regions are segmented into three methylation states as well. Comparing the identified regions with the result from the whole genome bisulfite sequencing (WGBS) datasets segmented by MethySeekR, our method is verified. Genome-wide maps of chromatin states show that methylation state is inversely correlated with active histone marks. Genes regulated by un-methylated regions are expressed and regulated by full-methylated regions are repressed. Our method is illustrated to be useful and robust.
Our method is valuable for DNA methylation genome-wide analyses. It is focusing on identification of DNA methylation states on microarray methylation datasets. For the features of array datasets, using two layers two-state HMM to identify to methylation states on CpG sites and regions creatively, our method which takes into account the distribution of genome-wide methylation levels is more reasonable than segmentation with a fixed threshold.
DNA 中胞嘧啶碱基的甲基化是许多真核生物中重要的表观遗传标记，也与正常和病变细胞的发育和进展有关。因此，对整个基因组的 DNA 甲基化进行分析对于理解表观遗传的影响至关重要。近年来，Illumina HumanMethylation450（HM450K）和 MethylationEPIC（EPIC）BeadChip 已广泛用于分析人类样本中的 DNA 甲基化。基于微阵列甲基化数据集预测 DNA 区域甲基化状态的方法对于进行全基因组分析至关重要。
我们报告了一种基于两层两状态隐马尔可夫模型（HMM）的计算方法，用于识别 HM450K 和 EPIC BeadChip 中单个 CpG 位点和 DNA 区域的甲基化状态。使用该方法，HM450K 和 EPIC 在 H1-hESC 和 GM12878 细胞系中检测到的所有 CpG 均被鉴定为未甲基化、中甲基化和全甲基化状态。大量 DNA 区域也被分割成三种甲基化状态。将识别出的区域与 MethySeekR 分割的全基因组亚硫酸氢盐测序（WGBS）数据集的结果进行比较，验证了我们的方法。染色质状态的全基因组图谱表明，甲基化状态与活性组蛋白标记呈负相关。受非甲基化区域调控的基因表达，受全甲基化区域调控的基因被抑制。我们的方法被证明是有用和稳健的。
我们的方法对于全基因组 DNA 甲基化分析很有价值。它专注于识别微阵列甲基化数据集中的 DNA 甲基化状态。对于阵列数据集的特点，创造性地使用两层两状态 HMM 来识别 CpG 位点和区域的甲基化状态，我们的方法考虑了全基因组甲基化水平的分布，比使用固定阈值进行分割更合理。
