Fig. 5: Plasma methylome-based detection of BM development.

a, Flow diagram of the approach used to build and evaluate classifiers of BM versus other main brain tumors in a clinical differential diagnosis and non-CNS tumor controls. b, MDS plot using DMRs selected between each class (BM, glioma, CNSL and control) during model building, with BM plasma samples clustering separately from other entities. c, Ensemble of ROC curves evaluating 50 iterations of BM-versus-others classifiers in the validation dataset plasma, demonstrating reliable discrimination of BM from others, including glioma, CNSL and non-CNS tumor controls. d, Box plot of Spearman correlation coefficient values, showing alignment between methylation signatures in paired BM tumor and plasma samples (n = 25). e,f, Representative individual clinical cases of a BM (e) and a glioma (f) that were accurately identified as a BM and non-BM (that is, ‘other’), respectively, based on testing dataset class probabilities from the BM-versus-others cfMeDIP–seq models. In the box plots, the central bars correspond to the median; the upper and lower distribution quartiles are displayed using boxes; and the 1.5× interquartile range (IQR) is represented by the whiskers.