FIGURE 2.

Predictive accuracy of the omics profiles and their combinations when the maximum number of features was limited. The points are average AU-ROC values over the 10 CV rounds in nested CV. Example feature sets from omics data are shown at limits x = 2, 10, and 30. The gene lists contain the features that were selected by Lasso in all the 10-fold at that limit. Expression data and integrated data share most genes in common. Black color indicates genes that were selected by both gene expression and integrated data, the pink color those genes that were selected by integrated data only. Note that the top genes of the two types of integrated data are the same. No copy number features were selected in all the 10 CV folds (no robust CNV features). See Supplementary Figure 4 for the version of ROC and PRCs with SEMs included.