Abstract
Background Technological advances in high throughput DNA methylation microarrays have allowed dramatic growth of a new branch of epigenetic epidemiology. DNA methylation datasets are growing ever larger in terms of the number of samples profiled, the extent of genome coverage, and the number of studies being meta-analysed. Novel computational solutions are required to efficiently handle these data.
Methods We have developed meffil, an R package designed to quality control, normalize and perform epigenome-wide association studies (EWAS) efficiently on large samples of Illumina Infinium HumanMethylation450 and MethylationEPIC BeadChip microarrays. We tested meffil by applying it to 6000 450k microarrays generated from blood collected for two different datasets, Accessible Resource for Integrative Epigenomic Studies (ARIES) and The Genetics of Overweight Young Adults (GOYA) study.
Results A complete reimplementation of functional normalization minimizes computational memory requirements to 5% of that required by other R packages, without increasing running time. Incorporating fixed and random effects alongside functional normalization, and automated estimation of functional normalisation parameters reduces technical variation in DNA methylation levels, thus reducing false positive associations and improving power. We also demonstrate that the ability to normalize datasets distributed across physically different locations without sharing any biologically-based individual-level data may reduce heterogeneity in meta-analyses of epigenome-wide association studies. However, we show that when batch is perfectly confounded with cases and controls functional normalization is unable to prevent spurious associations.
Conclusions meffil is available online (https://github.com/perishky/meffil/) along with tutorials covering typical use cases.
