Fig. 1

Identification of prognostic somatic mutation gene signature. DNA-seq prognostic analysis and signature generation pipeline. TCGA somatic mutation data is randomly split into training and validation datasets. Univariate Cox analysis identifies mutated genes associated with survival and only significant genes (FDR ≤ 0.05) are considered further. Bidirectional stepwise model selection for multivariate Cox analysis is used to select the final prognostic somatic mutation gene signature. Risk scores for patients in both training and datasets are computed using the final signature. The 75% percentile risk score of the training dataset is used as the stratification threshold for the KM analysis on both the training and validation datasets