Fig. 9
Evaluation of CS-based machine learning models for predicting immunotherapy response. (A) Single-cell visualization of cell clusters in basal cell carcinoma (BCC) patients receiving anti-PD-1 immunotherapy (GSE123813). (B) Ratios of cell proportion before and after immunotherapy (top panel) and distribution of CS subgroups in each cell type (bottom panel). (C) Differences in the proportion of CS subgroups between responders and non-responders before and after immunotherapy. (D) The genetic algorithm-artificial neural network (GA-ANN) was used for feature selection. (E) The order of importance of CS-related cancer driver genes calculated using GA-ANN. (F) Comparison of multi-class AUCs estimated by four machine-learning models in the training and testing sets. (G–J) Application of CS-based machine-learning models in 19 immunotherapy cohorts. Top panel: expression of PD-1, PD-L1 and CTLA-4 in the CS subgroups. Bottom panel: (i) proportion of responders in each CS subgroup; (ii) proportion of CS subgroups in responders and non-responders before and after immunotherapy; (iii) Kaplan-Meier survival analysis of CS subgroups.
