Detection of differentially methylated regions in whole genome bisulfite sequencing data using local Getis-Ord statistics
Detection of differentially methylated regions in whole genome bisulfite sequencing data using local Getis-Ord statistics
复制标题
使用本地 Getis-Ord 统计数据检测全基因组亚硫酸氢盐测序数据中的差异甲基化区域。
DOI：
10.1093/bioinformatics/btw497
复制
发表时间：
2016-11-15
期刊：
影响因子：
5.8
通讯作者：
Li, Zhiguang
中科院分区：
文献类型：
--
作者：
Wen, Yalu;Chen, Fushun;Li, Zhiguang
动机：DNA甲基化是一种重要的表观遗传修饰，在基因调控、细胞分化和癌症发生中起着重要作用。亚硫酸氢盐测序是一种广泛使用的获得全基因组DNA甲基化谱的技术，而分析亚硫酸氢盐测序数据的关键任务之一是检测不同处理条件下样品之间的差异甲基化区域（DMR）。虽然已经提出了许多工具来检测样品之间的差异甲基化的单CpG位点（DMC），直接DMR检测，特别是复杂的研究designs.Results的方法，在很大程度上是有限的：我们提出了一个新的软件，GetisDMR，直接DMR检测。我们使用β-二项回归对全基因组亚硫酸氢盐测序数据进行建模，其中甲基化水平的变化和混杂效应已经被考虑在内。我们采用了区域明智的测试统计量，这是来自当地的Getis-Ord统计，并考虑附近的CpG位点之间的空间相关性，检测DMR。与现有的方法，试图推断DMR从DMC的经验标准的基础上，我们提供统计推断直接DMR检测。通过广泛的模拟和两个小鼠数据集的应用，我们证明GetisDMR实现了更好的灵敏度，阳性预测值，更准确的位置和更好的协议DMR与当前的生物学知识。
Motivation: DNA methylation is an important epigenetic modification that has essential role in gene regulation, cell differentiation and cancer development. Bisulfite sequencing is a widely used technique to obtain genome-wide DNA methylation profiles, and one of the key tasks of analyzing bisulfite sequencing data is to detect differentially methylated regions (DMRs) among samples under different treatment conditions. Although numerous tools have been proposed to detect differentially methylated single CpG site (DMC) between samples, methods for direct DMR detection, especially for complex study designs, are largely limited.Results: We present a new software, GetisDMR, for direct DMR detection. We use beta-binomial regression to model the whole-genome bisulfite sequencing data, where variations in methylation levels and confounding effects have been accounted for. We employ a region-wise test statistic, which is derived from local Getis-Ord statistics and considers the spatial correlation between nearby CpG sites, to detect DMRs. Unlike existing methods, that attempt to infer DMRs from DMCs based on empirical criteria, we provide statistical inference for direct DMR detection. Through extensive simulations and an application to two mouse datasets, we demonstrate that GetisDMR achieves better sensitivities, positive predictive values, more exact locations and better agreement of DMRs with current biological knowledge.