methylation-data-clustering

methylation-data-clustering

Summary

Hierarchical clustering and principal component analysis of methylation profiles across multiple samples to assess sample relationships and methylation-based similarity. This skill reveals grouping patterns and variance structure in DNA methylation data, enabling identification of sample subpopulations and quality assessment of methylation profiling.

When to use

After merging methylation call files across all samples into a unified methylBase object (via unite()), when you need to assess whether biological replicates cluster together, identify unexpected sample groupings, or visualize global methylation similarity relationships before proceeding to differential methylation analysis.

When NOT to use

• Input is a single sample or fewer than 2 samples — clustering requires multiple samples to show relationships.
• methylBase object has not been constructed via unite() with consistent coverage filtering — inconsistent base coverage across samples will distort distance metrics.
• Samples have extremely divergent methylation profiles (e.g., from different organisms or methylation assay types) — correlation-based distance may not be meaningful.

Inputs

• methylBase object (unified methylation matrix from unite() with all samples merged at common coverage positions)
• Sample annotation table or experimental design metadata (optional, for color-coding clusters)

Outputs

• Dendrogram object (hierarchical clustering tree with samples as leaves)
• Scree plot (variance explained by each principal component)
• PC1/PC2 scatter plot (sample positions in principal component space)
• Clustering and PCA visualizations suitable for publication or QC reporting

How to apply

Load the methylBase object produced by unite() and apply clusterSamples() to perform hierarchical clustering on methylation profiles using correlation distance with Ward linkage, generating a dendrogram that visualizes sample relationships. Simultaneously apply PCASamples() to compute principal components and generate a scree plot showing variance explained by each PC. Extract and visualize the first two principal components (PC1 and PC2) as a scatter plot to assess sample grouping and methylation-based similarity. Interpret the dendrogram branch structure and PC1/PC2 loadings to confirm that biological/treatment replicates cluster together and that the methylation variance structure aligns with experimental design expectations.