Fig. 6: Application of minimalistic classifier methodology to methyl-seq data.
From: Minimalist approaches to cancer tissue-of-origin classification by DNA methylation
We trained a classifier using data from TCGA and FFPE methylation array cases. This classifier was tested on 15 institutional FFPE primary cases, for which both methyl-seq (triangles) and EPIC array (circles) data were available (top panel). The reference cancer type is color coded in the sample name labels on the X-axis and the predictions of the classifier are color coded (by cancer type) and plotted with respect to their levels of confidence. The classifier correctly predicted 11 out of 15 diagnoses (73.3%; of which, 10 of 15 had medium or high levels of confidence) based on methyl-seq data, and 14 out of 15 diagnoses (93.3%) based on array data. Below each sample, the correlations between methylation levels across the 24 probes for the two platforms are shown (bottom panel). BLCA bladder carcinoma, CESC cervical and endocervical cancers, CORE colorectal adenocarcinoma, LUAD lung adenocarcinoma, SKCM skin cutaneous melanoma, UCEC uterine corpus endometrial carcinoma.
