Fig. 3

Constructing the prognostic circRNA model in the RNA-seq dataset. a Univariate Cox analysis of the circRNAs expression with prognosis in RNA-seq dataset. The error rate for the data in random survival forests analysis as a function of trees and according to important score to filter Top 300 (b) or 200 (c) high expression of prognostic circRNAs. d Flow chart for identification of existing circRNA. e Agarose electrophoresis of selected circRNAs PCR products. f The results of Sanger sequence for the four circRNAs in the prognostic signature. g Screening out the multi-circRNA signature with largest AUC from all 512 signatures which were calculated by ROC for k = 1, 2…… 9 in the plot