illumina-methylation-array-preprocessing

Illumina methylation array preprocessing

Summary

Preprocessing and quality control of Illumina methylation array data (450K and EPIC) via detection p-value and bead count filtering, followed by type-2 probe correction and normalization. This skill removes low-quality probes and corrects technical biases prior to downstream analysis.

When to use

Your input is raw .idat files or a beta-valued matrix from Illumina HumanMethylation450 or EPIC arrays, and you need to remove unreliable probes (those with detection p-value > 0.01 or insufficient bead counts) before performing differential methylation or other downstream analyses. Apply this skill early in the pipeline when data quality and probe reliability directly affect downstream statistical power.

When NOT to use

• Input is already a curated, pre-normalized feature matrix from a published study without raw .idat or detection p-value metadata — skip to downstream analysis.
• Your array platform is not Illumina 450K or EPIC (e.g., custom or non-Illumina bisulfite sequencing data) — choose a platform-specific preprocessing pipeline.
• You have already applied aggressive probe filtering (< 5,000 probes remaining) — additional filtering may eliminate too much data to support statistical inference.

Inputs

• .idat files (raw Illumina methylation intensity data)
• beta-valued matrix (pre-imported methylation beta values)
• HumanMethylation450 or EPIC array sample metadata (sample IDs, phenotypes)

Outputs

• Filtered and normalized beta-value matrix
• Quality control report (probes retained vs. removed, detection p-value and bead count distributions)
• Corrected type-2 probe intensity values
• Batch-effect diagnostics (optional SVD visualizations)

How to apply

Load the raw methylation array data using ChAMP's data import functions (from .idat files or a beta-valued matrix). Apply champ.filter() with default parameters to execute two successive filtering steps: (1) removal of probes with detection p-value > 0.01, indicating unreliable signal, and (2) removal of probes with fewer than 3 beads in at least 5% of samples, which reflects insufficient technical replication. After filtering, apply type-2 probe correction using SWAN, PBC, or BMIQ (default is BMIQ) to adjust for the distinct hybridization kinetics of Infinium type-2 probes. Finally, apply Functional Normalization or another normalization method to correct for technical variation while preserving biological signal. Document probe retention rates and quality metrics (detection p-value distributions, bead count distributions pre- and post-filtering) to validate the filtering and correction steps.