The Illumina Infinium HumanMethylation450 Beadchips have been widely utilized in epigenome-wide association studies (EWAS). However, the existing two types of probes (type I and type II), with the distribution of measurements of probes and dynamic range different, may bias downstream analyses. Here, we propose a method, MGMIN (M-values Gaussian-MIxture Normalization), to correct the probe designs based on M-values of DNA methylation. Our strategy includes fitting Gaussian mixture distributions to type I and type II probes separately, a transformation of M-values into quantiles and finally a dilation transformation based on M-values of DNA methylation to maintain the continuity of the data. Our method is validated on several public datasets on reducing probe design bias, reducing the technical variation and improving the ability to find biologically differential methylation signals. The results show that MGMIN achieves competitive performances compared to BMIQ which is a well-known normalization method on β-values of DNA methylation.

Illumina Infinium HumanMethylation450 Beadchips已广泛用于表观基因组关联研究（EWAS）。但是，现有的两种类型的探头（I型和II型）的探头测量分布和动态范围不同，可能会影响下游分析。在这里，我们提出一种方法MGMIN（中号值高斯混合标准化），以根据 中号-DNA甲基化值。我们的策略包括分别将高斯混合分布拟合到I型和II型探头，中号值转换为分位数，最后基于 中号-DNA甲基化的-值以保持数据的连续性。我们的方法在减少探针设计偏差，减少技术差异和提高发现生物差异甲基化信号的能力的几个公共数据集上得到了验证。结果表明，与BMIQ相比，MGMIN具有竞争性能。BMIQ是众所周知的DNA甲基化β值归一化方法。