Fig. 1

PMLs show transcriptional heterogeneity. A Heatmap of top 2000 highly varying genes, with rows (genes) and columns (patients) sorted by hierarchical clustering. The columns are split into three groups as determined by a 3-way cut of the dendrogram. Row annotation highlights epithelium-related genes in red and immune-related genes in green. B (top) UMAP analysis of gene expression profiles from all samples colored by histopathological group. The progression status of 27 patients is denoted with black circles (stable), blue squares (progressed to dysplasia), and red squares (progressed to SCC). B (bottom) Clustering analysis using K-means, with k = 3. Within each cluster, the group positively enriched by chi-square test is highlighted. C Boxplots of GSVA-based enrichment scores. Displayed from left to right are: score differences of up- and down-regulated OPMD signatures (n = 37 genes; up = 9, down = 27); score differences of up- and down-regulated TCGA HNSCC signatures (n = 2647; up = 1317, down = 1330); and scores from a unidirectional p-EMT signature (n = 100). P-values are obtained using an ANOVA test