Fig. 5

sysSVM2 predictions in 34 cancer types. a Number of damaged canonical drivers per sample. Lists of canonical drivers for each cancer type were obtained from NCG [11] and mapped to samples of the corresponding cancer type. Six thousand sixty-seven samples with less than five canonical drivers damaged underwent the top-up procedure to reach five drivers. Difference in areas under the curve (AUCs) between the pan-cancer and cancer-specific settings in ranking canonical drivers over the rest of human genes and false positives (b) and in the composition score of the top five predictions (c). The median values of the distributions in each cancer type were used for comparison, with the yellow and blue regions indicating better performance in the pan-cancer and cancer-specific settings, respectively. The number of samples used for training is indicated on the x-axis. Colour dots represent cancer types where the two settings differ both significantly (FDR < 0.05, Wilcoxon rank-sum test) and substantially (|difference in medians| > 0.05 for AUCs, > 1 for composition score). ACC, adrenocortical carcinoma; TGCT, testicular germ cell tumours; PAAD, pancreatic adenocarcinoma; READ, rectum adenocarcinoma; MESO, mesothelioma; UVM, uveal melanoma; and OSCC, oesophageal squamous cell carcinoma. d Recurrence of damaging alterations in 282 canonical driver genes and 4470 sysSVM2 top-up predictions across 7646 samples. e Gene set enrichment analysis of sysSVM2 top-up genes, grouped in broad biological processes (Reactome level 1). Numbers of pathways enriched in at least one cancer type out of the total pathways tested are reported in brackets. Red vertical strokes indicate the mean number of cancer types that pathways from each broad process are enriched in (bottom x-axis)