Fig. 2

Cell–cell communication networks, functional annotation, and survival analysis a. An overview of the BulkSignalR model workflow for predicting ligand-receptor interaction. For a ligand-receptor interaction pair to be considered active, the corresponding downstream signalling targets must have a significantly correlated expression level (left panel). Selected ‘active’ ligand-receptor pairs and the enriched downstream signalling pathways are displayed (right panel). CR, complete remission; PD, progressive diseases; PR, partial remission . b. Functional annotation of top-ranked LDA genes for complete remission (left), partial remission (middle), and progressive disease (right). Functional annotation was performed by querying the Reactome database for overrepresented pathways. Pathways were ordered by significance level (p-value), blue to red indicates lowest to highest p-values. c. Boxplots displaying the expression levels of some doublet-specific genes (MCM3 and TACSTD2) which were significantly associated with clinical prognosis in both univariate and multivariate Cox regression models. A Kruskal–Wallis test was performed to compare expression across all groups while Wilcoxon tests were performed for pairwise comparisons between/within doublets and singlets. NS: non-significant; *: p < 0.05; **: p < 0.01; ***: p < 0.001. d. Venn diagram depicting the number of genes that were significantly associated with clinical prognosis following Cox regression models of deconvoluted gene expressions for the cancer, stromal and immune cell compartments. 21 genes (listed out) were found to be common to the 3 compartments and 17 of these were found to be significantly associated with clinical prognosis in the bulk (non-deconvoluted) data.