Fig. 2: Revisiting malignant state heterogeneity.

a, Jaccard similarity indices of the robust NMF programs (n = 707) based on their top 50 genes. Programs are ordered according to clustering and grouped into MPs (Methods). MP10 (stress1) and MP15 (stress2) were coalesced into a single stress state. MP9 (excitatory neuron (ExN)) and MP14 (neurogranin neuron) were coalesced into a single NEU-like state. The ‘cycling’ MP is separated to emphasize it as a feature that cells can have on top of their neurodevelopmental identity (and not as an independent state; Extended Data Fig. 3a). b, Gene expression scores of each cell for the different cellular states and the cell cycle. Cells are aggregated according to the state to which they were classified and are ordered within each group according to their cell cycle score. ‘Unresolved’ represents cells that could not be confidently classified to a particular state. c, Relative expression (centered log₂) of the three new MPs discovered in this work. Each row represents a gene and each column a cell. d, Association between the malignant MPs (rows) and gene sets reflecting normal brain development and adult brain cell types (columns). Each cell in the heatmap represents the Pearson correlation coefficient between the gene expression scores of an MP and a normal brain gene set across the malignant cells in the dataset. e, Association between pathway-based (rows) and gene-based (columns) MPs. Each cell in the heatmap represents the Pearson correlation coefficient between the gene expression scores of the pathway-based and gene-based MPs. The colored squares represent the seven groups of PMPs and their associated MPs. f, Demonstration of the hybrid state concept. The rows represent genes, which are aggregated according to the MP to which they belong. The columns represent cells, which are aggregated according to their singular (AC, GPC) or hybrid (AC-GPC) state. The heatmap of genes’ centered log₂ expression shows that hybrids express two MPs to the same extent, whereas singulars highly express a single MP. g, GBM cellular hierarchy model stemming from the high-frequency hybrid pairs. Each vertex represents a state; vertices are connected when they represent a high-frequency hybrid state. h, Two-dimensional representation of cellular states. Each quadrant reflects a cellular state and each dot represents a malignant cell; exact cell positions reflect their relative scores for the MPs, and their colors reflect the proportion of cells around them classified to the GPC state.