59 common differentially expressed FRGs were screened out. GO/KEGG enrichment and PPI network were executed. Based on 16 prognostic related FRGs identified by univariate Cox regression in GSE17674, 2 molecular clusters were screened out via NMF consensus. Survival rate and immune infiltration were totally different in two clusters. Subsequently multivariate/step Cox regression was conducted to identify 7 risk signatures (SLC2A1, PCK2, CHAC1, ATG13, PRKAA2, ARNT, SIRT1). K-M survival (p = 1.785e-06) and ROC curves (with AUC value 0.816, 0877, 0.919 in 1, 3, 5 years) were plotted to assess the good predictive ability of risk model. ICGC dataset with K-M survival (p = 1.558e-02) and AUC value (0.886, 0.750, 0.709 in 1, 3, 5 years) was used to validate the risk model. Risk score and clinical features (gender, age stage status) were incorporated into a nomogram model. Immune microenvironment (IME) ingredients (ESTIMATE score, immune cells, immune related pathways, and checkpoint genes) between two risk groups were also explored. High risk group possessed an activated immune statue compared to low risk group. Finally, prognostic signatures exhibited perfect diagnostic ability in ES occurrence, and several drugs showed IC50 sensitivity to different risk groups.