Additional file 16 of Systematic interrogation of mutation groupings reveals divergent downstream expression programs within key cancer genes

Additional file 16: Figure S15. Comparing subgroupings against mutation subsets defined by other tools for measuring variant significance. Classification tasks were created in which the top n samples according to the value of various continuous mutation properties were treated as a discrete subgrouping. Using the same training and testing regime as before, we compare the AUCs for these tasks to those for subgrouping tasks created using our original discrete approach. This revealed cases such as EGFR in TCGA-LUAD and NFE2L2 in TCGA-HNSC(HPV-) where using subgroupings was clearly superior to using these metric cutoffs as well as cases such as TP53 and PIK3CA in TCGA-BRCA(LumA) where neither subgroupings nor cutoffs significantly outperform the gene-wide classifier. Legend labels are annotated with an asterisk for classes of subgroupings in which the best subgrouping was cv-significantly better than the original gene-wide task. We include here these figures for the four cases listed above and for the remaining genes and cohorts at our data portal under Figures/S15 - Threshold Subgrouping AUC Comparisons. The names of these figures have the format (cohort)__(gene)__sub-comparison_Ridge.svg.