Fig. 3: Unsupervised clustering of probes by their usage.
From: Explainable artificial intelligence of DNA methylation-based brain tumor diagnostics
a t-SNE dimensionality reduction and unsupervised clustering of 10,000 selected probes according to their usage (Pearson’s distance). A total of 88 clusters are identified. Individual clusters of probes are numbered. b Heatmap shows the total probe usage by class and by cluster. Blue color indicates clusters in which probes are predominantly hypomethylated in a given tumor class, red color indicates clusters which are predominantly hypermethylated. Clusters (columns) and tumor classes (rows) are ordered by hierarchical clustering. Values are scaled by column. c t-SNE as in panel (a) showing the probe usage in ETMR compared to all other classes. The color code indicates hypomethylated (blue) and hypermethylated (red) probes. d Scatterplot showing the usage of probes belonging to cluster 78 (n = 44) for each tumor class. e t-SNE as in panel (a) showing the probe usage in A IDH compared to all other classes. f Scatterplot showing the usage of probes belonging to cluster 1. g Karyotype plot showing the genomic location of the probes belonging to cluster 78. h Similar karyotype plot showing probes belonging to cluster 1. Source data are provided as a Source Data file.
