Fig. 2
From: PathExpSurv: pathway expansion for explainable survival analysis and disease gene discovery
a Performance comparison on Prior Net, Fully-connected Net and PathExpSurv. Generally, the Fully-connected Net and PathExpSurv outperformed the Prior Net. On the THCA dataset, PathExpSurv even showed better result than the Fully-connected Net which had more learnable parameters. b GSEA p-values of the ranked genes list for each pathway. The GSEA p-values of PathExpSurv are significantly smaller than those of Fully-connected Net, indicating PathExpSurv has the ability to obtain meaningful expanded pathways and the results is more interpretable. c Example of training curves of the two-phase training. The loss and C-index showed significant improvement in the training phase. d Performance comparison on several methods of cancer survival analysis. The C-index results of 6 methods (Cox regression, Elastic-Net Cox model, Random Survival Forest, DeepSurv, DeepOmix and PathExpSurv) are shown, and PathExpSurv had best performance among these methods