Epigenetic classification of CLL using Me-iPLEX. (A) Heat map identifying the most differentially methylated CpGs among CLL epitypes and between CLL and normal cell types using Illumina 450K arrays. The 500 most variable 450K probes among the 3 CLL epitypes (top row). The 100 most differentially methylated probes comparing CLL with normal cells, myeloid cells, or T/NK cells (lower rows). Columns represent CLL patients (n = 310), normal myeloid cells (n = 36), normal T/NK cells (n = 12), and normal PBMCs (n = 20). (B) Density plots showing the distribution of the raw random forest score (upper panel) and the calibrated probability (lower panel) of each methylation class in our training sample set (n = 139). (C) Unsupervised clustering using a t-SNE plot of the training cohort samples. Colors represent the calibrated subtype call. (D) Confusion matrix comparing the known epitype call (determined from Illumina 450K analysis) with the epitype call by the random forest method of the training cohort (upper panel) and with the validation cohort (lower panel, n = 149).