Predictive accuracy using data types other than whole-genome sequencing.

(A) ROC curves for simulated whole-exome sequencing data, for one cancer type versus all others. Area under the ROC curve (mean across crossvalidation folds) for each cancer type is reported in the legend. (B) Crossvalidation AUPRC score for classifiers trained on various feature types under conditions of false negative mutation calls. (C) Crossvalidation AUPRC score for copy number states in 1Mb segments, estimated from SNP array data and tentatively divided into driver and passenger events. In (B) and (C), mean ± standard deviations are shown over multiple crossvalidation runs; (B) depicts the median score across all cancer types, and mean and s.d. thereof across crossvalidation runs are shown.