Detecting aberrant DNA methylation in Illumina DNA methylation arrays: a toolbox and recommendations for its use

In this study, our goal was to determine probe-specific thresholds for identifying aberrant, or outlying, DNA methylation and to provide guidance on the relative merits of using continuous or outlier methylation data. To construct a reference database, we downloaded Illumina Human 450K array data for more than 2,000 normal samples, characterized the distribution of DNA methylation and derived probe-specific thresholds for identifying aberrations. We made the decision to restrict our reference database to solid normal tissue and morphologically normal tissue found adjacent to solid tumours, excluding blood which has very distinctive patterns of DNA methylation. Next, we explored the utility of our outlier thresholds in several analyses that are commonly performed on DNA methylation data. Outliers are as effective as the full continuous dataset for simple tasks, like distinguishing tumour tissue from normal, but becomes less useful as the complexity of the problem increases. We developed an R package called OutlierMeth containing our thresholds, as well as functions for applying them to data.

Automated summary

The aim of this study was to determine probe-specific thresholds for identifying aberrant DNA methylation and provide guidance on using continuous or outlier methylation data. A reference database was constructed using Illumina Human 450K array data for over 2,000 normal samples, and probe-specific thresholds for identifying aberrations were derived. The utility of the outlier thresholds was explored in various analyses, and an R package called OutlierMeth containing these thresholds was developed.