Skip to main content
. 2014 Aug 22;30(17):i549–i555. doi: 10.1093/bioinformatics/btu467

Fig. 2.

Fig. 2.

Classification of 200 (HCD list) and 144 (Cancer5000 list) cancer driver genes into the classes Act and LoF. The training set of OncodriveROLE constitutes of all ‘Dom’ and ‘Rec’ labeled data points. ‘Dom?’ are CGC-annotated dominant genes excluded from the training set because of strong resemblance to the ‘Rec’ genes and previous literature evidence of this role. ‘DomT’ genes are CGC-annotated dominant genes only citing translocation events as prove and therefore not included in the training set. All ‘-’ labeled data points are driver genes not annotated in CGC, and whose prediction was the main goal of the study. The thresholds are drawn at 0.3 (as top limit of the LoF class) and 0.7 (as bottom limit of the Act class). Working with classification score thresholds of 0.3 (as top limit of the LoF class) and 0.7 (as bottom limit of the Act class), we classified 109 genes as LoF, 76 as Activating and left 15 genes as unclassified in the HCD list; meanwhile, we classified 97 genes as LoF, 43 as Activating and left 4 genes as unclassified (Fig. 2) in the Cancer5000 list. Genes for which we have observed <12 mutations were directly classified as ‘No class’ and assigned NA values in the classifications results (see Supplementary Tables S4 and S6)