Fig. 6

The ligand (L)- and receptor (R)-based risk model LRscore is a good prognostic and immune predictor. A Forest plot showing the hazard ratio (HR) of selected L/R pairs based on stepwise Cox regression analysis. B Kaplan–Meier curve showing that a high LRscore was associated with poor OS in TCGA-BLCA patients. C Radar chart showing that the LRscore was significantly correlated with risk according to univariate and multivariate Cox regression analyses. D The LRscore had the highest AUC and stability for predicting 1-, 3-, 5-, 7-, and 10-year survival. E Kaplan–Meier curves showing that a high LRscore was associated with poor DSS, DFS and PFS in TCGA-BLCA patients. F Kaplan–Meier curve showing that a high LRscore was associated with poor OS in IMvigor210 patients. G Stacked histogram showing that the high-risk group had a lower percentage of responding patients than the low-risk group. H The LRscore was negatively correlated with TMB and TNB in the anti-PD-L1 treatment cohort. I Three model genes demonstrated differential expression between the response (R) and nonresponse (NR) groups in the anti-PD-L1 treatment cohort. Wilcoxon test; *P < 0.05; **P < 0.01; ***P < 0.001. J Kaplan–Meier curves showing that ITGA6, HSPG2, LAMB1 and SDC2 are associated with the OS of IMvigor210 patients. K Receiver operator characteristic (ROC) curve showing the area under the curve (AUC) for different indicators used to identify patients with R or NR