dmr-detection-bumphunter

dmr-detection-bumphunter

Summary

Detect differentially methylated regions (DMRs) in EPIC or 450k methylation array data using the bumphunter-based method implemented in ChAMP. This skill identifies contiguous genomic regions with coordinated methylation differences between sample groups, filtering out single-CpG or low-complexity signals.

When to use

Apply this skill when you have preprocessed, normalized beta-value matrices from EPIC or 450k methylation arrays with at least two sample groups (case/control, treatment/untreated, or similar contrasts) and seek to identify regions of coordinated differential methylation rather than individual CpG sites. Use when you expect DMRs to contain multiple CpGs (≥3) and want bumphunter's spatial clustering approach rather than probe-level detection.

When NOT to use

• Input is already a list of called CpG-level differential methylation results; use DMR detection on raw or minimally processed beta values instead.
• Sample groups are not well-defined or biological replicates are absent; bumphunter requires sufficient statistical power within groups.
• Data contains strong batch effects or confounding variables not corrected during preprocessing; apply ComBat or RefbaseEWAS adjustment before DMR calling.

Inputs

• EPIC or 450k methylation array beta-value matrix (samples × CpG sites)
• Sample phenotype/group labels (case/control or multi-group contrast)
• Preprocessed, normalized methylation data (quality control and batch correction applied)

Outputs

• DMR object containing genomic coordinates, CpG membership, and statistics for each detected region
• DMR count and summary table with region-level p-values or test statistics
• Annotated DMR list with gene associations

How to apply

Load your EPIC or 450k methylation dataset (from .idat files or preprocessed beta-value matrix) into R and ensure it is normalized using one of ChAMP's supported methods (SWAN, PBC, BMIQ, or Functional Normalization from minfi). Call champ.DMR() function specifying the bumphunter-based detection method. The function will perform spatial clustering of CpGs and apply a minimum CpG threshold (typically ≥3 CpGs per region) to filter out isolated signals. Extract and count DMRs from the output; expect fewer regions than simulated DMRs due to filtering of single-to-dual-CpG signals. Verify DMR count is reasonable for your data scale and biological context (e.g., ~4700 DMRs for EPIC simulation data with 5000+ simulated regions indicates proper filtering).