sample-similarity-assessment-from-methylation

Sample similarity assessment from methylation

Summary

Assess methylation-based relationships among samples by computing hierarchical clustering and principal component analysis (PCA) on methylation profiles. This skill reveals sample grouping patterns and similarity structures that reflect methylation heterogeneity across treatment or control conditions.

When to use

After merging methylation calls across all samples using unite() to create a methylBase object, apply this skill to characterize whether replicate samples cluster together and to visualize methylation-driven separation between biological groups (e.g., test vs. control). Use it as an exploratory quality-control step before differential methylation analysis.

When NOT to use

• Sample methylation calls have not yet been merged across all samples (unite() has not been run); use methRead() and filtering steps first.
• Only single-sample or unpaired data is available; clustering and PCA require ≥2 samples for meaningful comparison.
• Samples have extremely low or unequal coverage after filtering; low-depth regions introduce noise into correlation and PC estimates.

Inputs

• methylBase object (unified methylation calls across all samples, produced by unite())

Outputs

• Dendrogram (hierarchical clustering tree with sample labels and correlation distances)
• Scree plot (variance explained by each principal component)
• PC1/PC2 scatter plot (2D sample projection with methylation-based similarity geometry)

How to apply

Load the merged methylBase object from unite() into methylKit. Apply clusterSamples() to perform hierarchical clustering using correlation distance with Ward linkage, which groups samples by methylation profile similarity and produces a dendrogram. Simultaneously apply PCASamples() to compute principal components and generate a scree plot showing variance explained by each component. Extract and scatter-plot the first two principal components (PC1 and PC2) to assess sample grouping in reduced dimensionality space. Samples that cluster together or occupy the same region in PC space indicate methylation-level agreement; separation indicates distinct methylation profiles. Examine both outputs together: the dendrogram confirms hierarchical relationships, while the PCA scatter plot validates separation in the top variance axes.