Fig. 7.

Identification of two TGF-β subtypes based on the deep neural network (DNN) associated with TGF-β signatures. a Schematic diagram of a deep neural network for predicting TGF-β subtypes. b The performance of our DNN classifier on training and testing sets; Left: area under roc curve (AUC) of receiver operating characteristic (ROC) curve across each tumor. Right: ROC curve of the DNN classifier in stomach adenocarcinoma. c TGF-β score in different TGF-β subtypes for each other dataset classified by the DNN model. d–g Kaplan-survival curves (including overall survival and disease-free survival) for different TGF-β groups identified by DNN model in GSE62254 and GSE17536. h Gene-set enrichment analysis showed that TGF-β and EMT pathways were significantly enriched in different TGF-β groups identified by DNN model. *P < 0.05, **P < 0.01, ***P < 0.001