Fig. 2

Construction and validation of a PyroScore prognostic model. A Univariate Cox regression analyses of PRGs associated with TCFi. The top genes with a P value < 0.2 are presented. B Venn diagram shows that five machine learning algorithms screened nine genes for constructing the prognostic model. C Coefficients of nine genes identified by multivariate Cox regression analyses. D PyroScore distribution, prognostic status of each patient, and heatmaps of nine prognostic PRGs. E TCFi KM curves of PyroScore subgroups. F ROC curves present the predictive efficiency of the PyroScore. G, H PCA plot (G) and t-SNE plot (H) show the separation of PyroScore subgroups. I-L TCFi KM curves and the ROC curves of PyroScore subgroups in the testing set (I, J) and the entire cohort (K, L)