Fig. 5

Subtype-specific immunophenotype and therapeutic prediction. A Comprehensive heatmap of immune microenvironment signatures, immune-checkpoint genes, and 22 types of infiltrated immune cells between Cluster1 and Cluster2 subtypes. B Heatmap of Cluster1 and Cluster2 subtypes for response of anti-CTLA-4 and anti-PD-1 in Chen et al. study. The results manifested that Cluster1 subtype could be more sensitive to the PD-1 inhibitor. C Heatmap of Cluster1 and Cluster2 subtypes for response of PD-1 in Prat A et al. study. The results indicated that Cluster1 subtype could have more chance to achieve complete response for anti-PD-1 therapy. D Heatmap of Cluster1 and Cluster2 subtypes for response of PD-1 in Hugo et al. study. The results indicated that Cluster1 subtype is more promising to achieve complete response for anti-PD-1 therapy. E Significant drug-disease scores for UM patients in the Connectivity Map. The top five potential drugs are labeled. F The risk heatmap of deep learning (DL) signature in TCGA-UVM cohort. G Kaplan–Meier curve of survival probability between high- and low-score of DL-signature in TCGA-UVM cohort. H Kaplan–Meier curve of survival probability between high- and low-score of DL-signature in HX cohort. I Time independent ROC curves and corresponding AUC values for DL-signature in TCGA-UVM cohort. J Time independent ROC curves and corresponding AUC values for DL-signature in HX cohort