2.12. Planned data analysis

We will use linear regression to test the association of an expected approximately 850,000 CpG sites with T2D case‐control status (T2D status is the predictor). Modeling the CpG as the outcome has been adopted in many epigenetic studies to enable direct adjustment for technical variables in association analysis (e.g., an estimated mixture of cell proportion, batch). ^{16 , 17 , 18 , 19} While we will let descriptive statistics (chi‐square and t‐tests) guide our covariate adjustment, our current plans include covariate adjustment for precocious puberty, glycemic control, BMI, and medication use. We will consider additional adjustments for the methylation array, plate row, and column, similarly to prior consortia reports. ¹⁸ In addition to considering single CpG sites, we will investigate DMRs using ChAMP, which uses three different algorithms (1) the bumphunter package; (2) DMRcate; and (3) the probe lasso function to identify DMRs between case–control status. Each algorithm is slightly different but considers annotated genomic features and their corresponding local probe densities to call and test the significance of DMRs between cases and controls using normalized beta values. ^{20 , 21 , 22} We plan to calculate DMRs with a methylation cut‐off of 0.015 (corresponding to a 1.5% difference in beta values by T2D status). ²³