methylation-batch-effect-detection

methylation-batch-effect-detection

Summary

Detect and visualize batch effects in DNA methylation array data (450K and EPIC) using singular value decomposition (SVD) with Random Matrix Theory component capping. This skill identifies latent batch-related variability before normalization or correction.

When to use

Apply this skill when you have loaded a normalized beta-valued methylation matrix (e.g., from HumanMethylation450 or EPIC arrays) and need to assess whether technical batch effects or unknown latent variables are confounding your sample groups prior to downstream analysis (DMR detection, differential methylation testing).

When NOT to use

• Input is raw (non-normalized) .idat intensity files — use champ.load() and normalization steps first
• You have already corrected batch effects with ComBat or similar and now seek to validate correction success — use post-correction QC plots instead
• Your data contains fewer samples than components to extract — SVD will be uninformative

Inputs

• normalized beta-valued matrix from HumanMethylation450 or EPIC array data
• optional: sample metadata table with batch, treatment, and technical covariates

Outputs

• SVD component loadings matrix (samples × up to 20 components)
• explained variance per component
• scree plot and heatmap visualizations
• batch effect assessment summary

How to apply

Load a normalized beta matrix into R and call champ.SVD() to perform singular value decomposition analysis. The function automatically applies Random Matrix Theory to detect the number of latent components in the data. When Random Matrix Theory identifies more than 20 latent components, ChAMP's implementation caps the output to the top 20 principal components for interpretability. Inspect the returned component loadings and variance explained to identify which experimental variables (batch, sample type, technical replicate) correlate with the principal components. Components with strong correlation to known technical factors (batch ID, array position, processing date) indicate batch effects requiring correction via ComBat or similar methods before proceeding to differential analysis.