Fig. 3: Identification and molecular characteristics of RSS.

a Delta plots to select the optimal number of clusters. 3 clusters were found to be the optimum number of clusters. b Silhouette plots of the hierarchical clusters on the exploration dataset (n = 182). c PCA plot of the hierarchical clusters on the exploration dataset. d 5-Fold validation and ROC Curve plots of the RSS classifier model on the exploration dataset. Accuracies were calculated based on the one-vs-all approach for each group. In each fold, 90 + % accuracies were achieved. e Heatmap of the gene modules and distribution of RSS, CMS, CRIS, MSI and age. Rows represent the patients, RSS groups on the left and their clinical information on the right. Columns represent the median value of each gene module. The top of the column shows the biological/molecular characteristics of the modules. Heatmap representation based on all the rectal samples (n = 870, Supplementary Table 3) after the RSS classification is applied to the rest of the microarray and RNA-Seq samples. f Kaplan–Meier survival plots of the rectal specific subtypes when classification derived from pre-treatment samples. RSS1 represents the best disease-free survival group while RSS2 has the worst disease-free survival. The p-value (0.022) is based on the log-rank test.