Figure 1

Unsupervised clustering of the 56 glioblastomas. Unsupervised hierarchical clustering of the 56 GBMs. The heatmap was done with the 72 probe sets used in the centroid classifier that was able to generate the 3 groups identified through unsupervised clustering (Additional file 1 Table S2). Samples and genes were clustered using Ward's linkage and 1-Pearson correlation coefficient. For each probe set, the lowest and highest intensity values are displayed in blue and red, respectively. Treatment: C = First-line chemotherapy, R = Radiotherapy. Response: N = non-responder, Y = responder. Phillips = class according to Phillips et al.'s classification [8], blue P = Proneural, green P = Proliferative, M = Mesenchymal. Verhaak = class according to Verhaak et al.'s classification [10], N = Neural, C = Classical, M = Mesenchymal, P = Proneural. The GBMs were classified into three groups: one group enriched in EGFR-amplified patients (n = 19, blue cluster), one group characterized by a high level of expression of immune and extra-cellular matrix genes (n = 12, red cluster) and one group characterized by a higher of expression of neural genes (n = 25, green cluster). There was a significant overlap when using Phillips et al.'s classes (Fisher's exact test p-value = 1.7 × 10^-8) and a larger overlap with Verhaak et al.'s classes (Fisher's exact test p-value = 2.2 × 10^-14).