Extended Data Fig. 1: Validation of diffuse midline glioma cell states and master regulators in an independent single-cell RNA sequencing dataset.

a, UMAP projection of VIPER-inferred protein activity across 1,629 tumor cells from an independent cohort of eight pediatric diffuse midline glioma (DMG) patient samples, with cells colored by clusters identified using resolution-optimized Louvain clustering. b, UMAP from (a) with cells annotated in a supervised manner based on statistically significant enrichment of previously defined human DMG cell state master regulators (MRs) within the VIPER-inferred protein activity signature of each cell (one-sided Gene Set Enrichment Analysis (GSEA), FDR ≤ 0.05), demonstrating strong concordance between unsupervised clustering and supervised cell state assignment. c, Distribution of the proportion of cells assigned to each DMG cell state (y-axis) across the eight patient tumors (x-axis), showing that all cell states are represented within each tumor. d, Heatmap of viperSimilarity normalized enrichment scores (NES) quantifying concordance between cell state MRs identified in the initial six-patient discovery dataset (D1; 3,032 tumor cells) and the independent eight-patient validation dataset (D2; 1,629 tumor cells). Red indicates concordant enrichment; blue indicates inverse enrichment. e, UMAP projections of VIPER-inferred protein activity from an integrated analysis of 4,661 tumor cells across all 14 DMG patients from datasets D1 and D2. Resolution-optimized Louvain clustering identified eight clusters (top), which were subsequently assigned to the seven previously defined DMG cell states based on lineage-marker protein activity (bottom). f, Heatmap of Stouffer-integrated protein activity (NES) for oligodendrocyte precursor cell (OPC), oligodendrocyte (OC), and astrocyte (AC) lineage markers, as well as proliferative markers (rows), across the eight unsupervised clusters from the integrated analysis (columns), supporting cluster annotation. g, Heatmap of viperSimilarity enrichment scores (NES) quantifying concordance between cell state MRs across all 14 individual patient samples, shown as a color gradient from blue (inverse enrichment) to red (concordant enrichment), demonstrating highly conserved regulatory programs across tumors. Cell state annotations and patient identity are indicated by the horizontal and vertical bars. h, Heatmap of viperSimilarity enrichment scores (NES) quantifying concordance between cell state MRs inferred from DMG single-cell regulatory networks and those derived from the DMG bulk regulatory network. Red indicates concordant enrichment and blue indicates inverse enrichment.