Fig. 2

A Heatmap of estimated exposures obtained by fitting 19 COSMIC signatures to the STAD dataset. Genome samples are displayed as columns of the heatmap while COSMIC signatures are arranged as rows and estimated (log-transformed) exposure levels are shown by the colour scale. B Fuzzy clustering of samples according to estimated exposures, compared to known classifications by molecular profiles. Clusters were organized in columns and for each sample (row) the colour code indicates the membership grade of each cluster. Following the fuzzy clustering approach, a hierarchical clustering algorithm was applied to the membership grades (dendrogram at left), enabling better visualization of results and allowing to establish a relation to molecular sub-types and MSI status (annotation columns at right side). C p values found by the Kruskal–Wallis test for differences in exposures among the four sample groups. For comparison and display purposes, the p values were inverted and log-transformed. Box-plots of obtained scores are displayed and the significance cutoff of 0.05 is indicated by the red line. The labels at the x-axis correspond to the ID of each signature and, for those showing significant differences, the group characterized by higher exposure levels. D ROC curve of the exposure-based classification of samples according to their MSI status and related confusion matrix. E Kaplan–Meier curves showing the overall survival of STAD patients after stratification by the exposures obtained while fitting COSMIC signature SBS26. The displayed p value was found by application of the log-rank test for defined sample groups