Skip to main content
Fig. 2 | Genome Medicine

Fig. 2

From: The prognostic potential of alternative transcript isoforms across human tumors

Fig. 2

a Illustration of the blind test on unlabeled patients. Patients without annotated stage were predicted using the model of the corresponding tumor type, for each of the stage classes independently. Patients predicted as early or late were collected into two separate groups and tested for differences in survival. This test was performed for each stage class independently and only using tumor types that showed an association between stage and survival in the labeled patients (Table 2). b, c Survival (Kaplan–Meyer) plots associated with the test for M-models and N-models, respectively. They indicate the survival percentage (y-axis) versus survival in months (x-axis) based on the predicted stage on the unannotated samples using the classifier for each corresponding tumor type. The p value in each plot corresponds to the Cox regression between the two groups and HR indicates the hazard ratio. d Accuracies of the transcript isoform models (I) compared to the gene (G) and event (E) models. Accuracies are given as boxplots for the distribution of AUC values (y-axis) from a tenfold cross-validation for each tumor type (x-axis) for the M-, S-, N-, and T-models. Tumors for which stage data were missing are not shown (Table 1). e Survival (Kaplan–Meyer) plot of the early and late-stage predictions performed with the gene-based S-models on unannotated samples. The p value corresponds to the Cox regression between the two groups. HR hazard ratio

Back to article page