Measuring variability in Illumina MethylationEPIC microarrays

DNA methylation microarrays have become a widely used tool for investigating epigenetic modifications in various aspects of biomedical research. However, technical variability in methylation data poses challenges for downstream applications such as predictive modeling of health and disease. In this study, we measure the impact of common sources of technical variability in Illumina DNA methylation microarray data, with a specific focus on positional biases inherent within the microarray technology. By utilizing a dataset comprised of multiple, highly similar technical replicates, we identified a chamber number bias, with different chambers of the microarray exhibiting systematic differences in fluorescence intensities and their derived methylation beta values, which are only partially corrected for by existing preprocessing methods, and demonstrate that this positional bias can lead to false positive results during differential methylation testing. Additionally, our investigation identified outliers in low-level fluorescence data which might play a role in contributing to predictive error in computational models of health-relevant traits such as age. Whole blood from four human donors were measured with total of sixteen technical replicates, across multiple slides and chambers.

DNA甲基化微阵列（DNA methylation microarrays）已成为生物医学研究各领域探究表观遗传修饰（epigenetic modifications）的通用工具。然而，甲基化数据中的技术变异给健康与疾病预测建模等下游分析应用带来了挑战。本研究评估了Illumina DNA甲基化微阵列数据中常见技术变异来源的影响，并特别聚焦于该微阵列技术固有的位置偏差。本研究利用由多个高度相似的技术重复样本组成的数据集，发现了芯片孔室编号偏差现象：微阵列的不同孔室在荧光强度及其衍生的甲基化β值上存在系统性差异，而现有预处理方法仅能部分校正该偏差；同时证实，此类位置偏差会在差异甲基化检测过程中引发假阳性结果。此外，本研究还在低强度荧光数据中识别出异常值，此类异常值可能会对年龄等健康相关表型的计算模型预测误差产生影响。研究共采集4名人类供者的全血样本，并在多张芯片玻片的不同孔室中完成了总计16次技术重复检测。