Additional file 5 of Systematic interrogation of mutation groupings reveals divergent downstream expression programs within key cancer genes

Additional file 5: Figure S4. Subgrouping prediction tasks outperform cohort-specific random background prediction tasks. For each of the learning tasks completed in METABRIC(LumA) to predict the presence of an actual mutation or subgrouping of mutations, five additional tasks were performed based on predicting a simulated set of point mutations. These sets were constructed by randomly selecting a group of samples from the entire cohort of the same size as the group of samples affected by the original “real” mutation in the cohort. In addition to these cohort-specific random subgroupings, we also constructed five gene-specific random subgroupings for each task by selecting samples as above using only those samples carrying any point mutation of the gene associated with the task’s subgrouping. (left) We compare the AUCs of cohort-specific random subgroupings to the AUCs of the original subgroupings. (right) We compare the AUC of each original subgrouping to the best AUC across the gene-specific random subgroupings of the same number of “mutated” samples. We include here these figures for the METABRIC-(LumA) and TCGA-BRCA(LumA) cohorts and for the remaining cohorts at our data portal under Figures/S4 - Cohort Null Background AUC Comparisons. The names of these figures have the format (expr-source)__(cohort)__cohort-comparison_Ridge.svg.