Extended Data Figure 6: miRNA clusters and miRNA-gene/protein anti-correlations in cervical cancer.

a, Unsupervised clustering for miRNA profiles across 178 core-set tumour samples. Top to bottom: a normalized abundance heatmap for the fifty 5p or 3p strands that were highly ranked as differentially abundant by a SAMseq multiclass analysis; a silhouette width profile calculated from the consensus membership matrix; covariates with associated P values; and a summary table of the number of samples in each cluster. The scale bar shows row-scaled log₁₀[RPM + 1] normalized abundances. Blue triangles mark selected cancer-associated miRs that were both differentially abundant across the subtypes and abundant in at least one subtype. b, Subnetworks of potential targeting relationships for a subset of miRNAs, as significance-thresholded (FDR < 0.05) miRNA–mRNA and miRNA–protein anti-correlations that are supported by functionally validated publications. For genes (nodes), colour distinguishes those that are only present in mRNA data (grey) from those that are present in both mRNA and RPPA data (green). Edges represent anti-correlations, and colour distinguishes anti-correlations between miRNA and mRNA (blue); and miRNA and unphosphorylated protein (green). In the n = 178 core-set cohort, no correlations satisfying FDR < 0.05 were reported between miRNA and phosphorylated protein.