Fig. 4: Single-sample gene signature analysis of highly pure GBM tumour fractions reveal two biologically distinct clusters.

a Two clusters of samples, independent of histomorphologic niches, are revealed by hierarchical clustering of samples using the ssGSEA scores from 64 selected gene signatures. Teal represents palisading cells around necrosis (PAN) samples while orange represents cellular tumor (CT) samples. b Functional drivers associated with each cluster using psSubPathways software package. c Inverse correlation between the KRAS targets and MYC targets signatures in the protein dataset generated in this study (n = 34 samples); additional null hypothesis testing was performed by permuted linear model fit analysis (lmPerm R package, p = 0.024). The 95% confidence interval is shown as grey areas. d 3D scatter plot of ssGSEA values of KRAS targets, MYC targets, and hypoxia gene signatures using proteomics (this study). e Transcriptomics (Ivy Gap) data from histomorphologically-defined samples. f Functional profiling by GSEA of Ivy Gap CT samples for the groups of samples labeled as MYC targets-high (n = 9) vs KRAS targets-high (n = 5) while presenting low activation of the hypoxia signature. g ROC analysis of clustering based on KRAS_targets, MYC_targets, or HYPOXIA gene signatures compared to that using Verhaak et al. mesenchymal, proneural and classical gene sets using cut-off optimization with the Youden method. h Distribution of KRAS_targets and MYC_targets signature enrichments in GBM at the single-cell level (Richards et al.) where (i) Mitotic spindle enrichment is quantified across the three detected populations. j Cell migration process enrichment across the three detected populations. Source data are provided as a Source Data file.