Fig. 1

An overview of the workflow in this study. In details, a set of promising protein-protein interaction sub-modules were identified from the HPRD database. The gene-based expression profile was transformed into PPI-module-based profiles by the GSVA algorithm. Using Cox regression model and Random forest algorithm, we established a prediction model named 11-PPI-Mod in the training dataset. This signature was also validated in another independent validation dataset. In addition, in vitro experiments were performed to validate one of the predictive genes in terms of chemoresistance