Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array

Area of Science:

• Epigenetics
• Genomics
• Epidemiology

Background:

• Growing number of studies identify DNA methylation differences linked to complex phenotypes.
• Analytical challenges in epigenetic epidemiology require further exploration.
• Focus on statistical significance, linear regression appropriateness, and sample size adequacy.

Purpose of the Study:

• Quantify the multiple testing burden in DNA methylation studies.
• Propose a standard statistical significance threshold.
• Assess the suitability and potential bias of linear regression for DNA methylation data.
• Determine the required sample size for adequately powered association studies.

Main Methods:

• Utilized the Illumina EPIC array on the Understanding Society cohort (n=1175).
• Simulated null DNA methylation studies to determine the empirical p-value distribution.
• Assessed linear regression assumptions with DNA methylation data.
• Conducted power calculations for DNA methylation association studies.

Main Results:

• Established a 5% family-wise error rate significance threshold of 9x10^-8 for EPIC array studies.
• Found that DNA methylation data often violate linear regression's normal residual assumption.
• Demonstrated that this violation does not appear to bias analyses or increase false positives.
• Concluded that studies with approximately 1000 samples are adequately powered for most sites.

Conclusions:

• A significance threshold of P < 9x10^-8 is recommended for EPIC array DNA methylation studies.
• Linear regression is confirmed as a valid statistical method for DNA methylation analysis.
• Findings provide a framework for interpreting results in current and future epigenetic epidemiology studies.