Fig. 8.

a–c The IC50 values of six anti-cancer drugs were predicted by pRRophetic algorithm and were Z-score normalized. The difference between high- and low-CAF risk groups of each drug in a TCGA COAD/READ, b GSE39582 and c GSE17536 cohort was compared by Wilcoxon test (** P < 0.01, *** P < 0.001). d–f TIDE analysis for predicting the likelihood of clinical response to immune checkpoint inhibitors in d TCGA COAD/READ, e GSE39582 and f GSE17536 cohort. Patients with a lower CAF risk score are more likely to respond from immune therapy