Fig. 4

Random Forest predictions, trained on gene-expression data, of gene-mutation status for lung adenocarcinoma. We identified genes that had been mutated in at least ten tumor samples and that contained no mutation in other genes that had been mutated in at least ten samples. Then using cross validation, we evaluated how well the Random Forest classification algorithm could identify which gene was mutated in a given sample. The algorithm produced a probabilistic prediction for each gene (class), and we evaluated these predictions using the area under the receiver operating characteristic curve (AUROC). The x-axis labels indicate AUROC values for each gene. Relatively high AUROC values indicate that gene-expression levels are highly predictive of gene-mutation status and thus suggest that mutations in these genes exert a characteristic effect on gene-expression levels