Critical evaluation of the reliability of DNA methylation probes on the Illumina MethylationEPIC BeadChip microarrays for dementia research.

Background DNA methylation (DNAm) has been implicated in many diseases including dementia. Array-based technologies offer a cost-effective and comprehensive approach for measuring DNAm on a genome-wide scale. However, the accuracy of DNAm measurements obtained using Illumina arrays can vary across different probes. Previous research has focused primarily on assessing the reliability of DNAm in younger subjects, and have compared duplicate samples between the 450k-450k or 450k-EPIC platforms, with limited investigations on EPIC-EPIC comparisons. Methods We conducted a comprehensive assessment of probe reliability on the Illumina EPIC arrays using 138 duplicated blood DNAm samples from subjects older than 65 years in the Alzheimer's Disease Neuroimaging Initiative (ADNI) study. To assess the reliability of each probe, we computed intraclass correlations (ICCs) for each probe. Both the magnitude and patterns of reliability in the EPIC-EPIC comparison were assessed. Furthermore, we also investigated the impact of probe reliability on the analyses of epigenome-wide association studies (EWAS). Results Our findings revealed the reliability of probes on the EPIC arrays is higher than those of previous studies involving duplicate measurements on 450k-EPIC or 450k-450k arrays. Consistent with earlier research, we observed increased reliability in probes with substantial between-subject variances or average methylation beta values ranging from 0.2 to 0.8. Lower reliability was observed in type I probes or probes located within the promoter and CpG island regions. In addition, we found some probes can yield high ICC values despite significant disagreement in duplicate measurements, primarily due to their relatively high between-subject variance. To account for such discrepancies explicitly, we introduced a novel statistical measure called the modified ICC, which penalizes the ICC based on the half-width of the 95% confidence limits of agreement. Importantly, we found probe reliability has significant implications in various downstream analyses of EWAS, such as meta-analysis, differentially methylated regions analysis, and integrative analyses within the cross-tissue or multi-omics contexts. Conclusion We developed a valuable resource for dementia research, providing crucial reliability information for probes on the EPIC array. This resource can be utilized to identify and prioritize high-quality probes, thereby minimizing the potential for false discoveries and maximizing the potential of EWAS.