Heiss, J. A. | bioRxiv Science

bioRxiv2022

Computational deconvolution of fifteen leukocyte subtypes from DNA methylation microarrays trained on flow cytometry data in the Health and Retirement Study

Adjusting for cell composition is critical in epigenome-wide association studies of whole blood samples. Using DNA methylation of whole blood samples (as opposed to purified cell types) and complete blood counts/flow cytometry data from 2530 participants in the Health and Retirement Study, we trained and tested a computational model that extends the number of estimated leukocyte subtypes to fifteen compared to established models with six or seven cell types. Our model, which can be applied to both Illumina 450k and EPIC microarrays, explained a larger proportion of the observed variance in whole blood DNA methylation levels than popular reference-based cell deconvolution approaches, and vastly reduced the number of false-positive findings in a reanalysis of an epigenome-wide association study of chronological age.

bioinformatics↗

bioRxiv2018

Guidance on filtering DNA methylation microarray probes by detection p-values

DNA methylation microarrays are popular for epigenome-wide association studies, but spurious outlier values complicate downstream analysis. It was demonstrated previously that commonly used detection p-value cut-offs were insufficient and led to many calls on the Y-chromosome probes in females. We extend these observations by assessing 2,578 samples from 18 studies run on the 450K chip as well as outliers across technical replicates. We provide comprehensive guidance and software for filtering methylation microarrays as an essential step to reduce outliers.

bioinformatics↗