Purity adjustment and the methylation state of the non-tumor compartment from bulk breast cancer data.

A) Heatmap clustering of the top 5000 most varying CpGs in the TCGA 630-sample BRCA cohort. Original methylation data (middle) with inferred non-tumor methylation (left) and purity-adjusted (right) data presented for comparison and contrast. B) Density plot showing effect of purity-adjustment on the beta-value distribution of the top 5000 most varying CpGs. C) Scatterplot of mean inferred beta versus mean of “normal breast” methylation beta values for the same 5000 CpG showing excellent overall correlation. D) Boxplot visualization of the profound effect of tumor purity when cutting a hierarchical clustering tree on the level of five clusters in unadjusted data. E) Boxplot visualization of the much-reduced effect of tumor purity on clustering in purity-adjusted data. F-G) Barplot visualization of the distribution of PAM50 subtypes by cluster in unadjusted (F) and adjusted (G) data showing more discrete partitioning of gene expression subtypes in adjusted clusters.