Recent studies on the role of epigenetics in disease have focused on DNA methylation (DNAm) profiled in bulk tissues limiting the detection of the cell type affected by disease-related changes. Advances in isolating homogeneous populations of cells now make it possible to identify DNAm differences associated with disease in specific cell types. Critically, these datasets will require a bespoke analytical framework that can characterize whether the difference affects multiple or is specific to a particular cell type. We take advantage of a large set of DNAm profiles (n = 751) obtained from five different purified cell populations isolated from human prefrontal cortex samples and evaluate the effects on study design, data preprocessing, and statistical analysis for cell-specific studies, particularly for scenarios where multiple cell types are included. We describe novel quality control metrics that confirm successful isolation of purified cell populations, which when included in standard preprocessing pipelines provide confidence in the dataset. Our power calculations show substantial gains in detecting differentially methylated positions for some purified cell populations compared to bulk tissue analyses, countering concerns regarding the feasibility of generating large enough sample sizes for informative epidemiological studies. In a simulation study, we evaluated different regression models finding that this choice impacts on the robustness of the results. These findings informed our proposed two-stage framework for association analyses. Overall, our results provide guidance for cell-specific epigenome-wide association studies, establishing standards for study design and analysis, while showcasing the potential of cell-specific DNAm analyses to reveal links between epigenetic dysregulation and disease.

MOTIVATION: The datasets generated by DNA methylation analyses are getting bigger. With the release of the HumanMethylationEPIC micro-array and datasets containing thousands of samples, analyses of these large datasets using R are becoming impractical due to large memory requirements. As a result there is an increasing need for computationally efficient methodologies to perform meaningful analysis on high dimensional data. RESULTS: Here we introduce the bigmelon R package, which provides a memory efficient workflow that enables users to perform the complex, large scale analyses required in epigenome wide association studies (EWAS) without the need for large RAM. Building on top of the CoreArray Genomic Data Structure file format and libraries packaged in the gdsfmt package, we provide a practical workflow that facilitates the reading-in, preprocessing, quality control and statistical analysis of DNA methylation data.We demonstrate the capabilities of the bigmelon package using a large dataset consisting of 1193 human blood samples from the Understanding Society: UK Household Longitudinal Study, assayed on the EPIC micro-array platform. AVAILABILITY AND IMPLEMENTATION: The bigmelon package is available on Bioconductor (http://bioconductor.org/packages/bigmelon/). The Understanding Society dataset is available at https://www.understandingsociety.ac.uk/about/health/data upon request. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.