Fig. 2.

Construction of a hierarchical clustering tree and modeling of risk prognosis. A–C Select appropriate soft thresholds in the GSE16515 dataset and perform average connectivity analysis for 1–30 soft-threshold powers. After creating a hierarchical clustering tree, correlation heatmaps and scatter plots show that the RED module strongly correlates with tumors. D–F Hierarchical clustering trees were created in the GSE16515 dataset in the same way as above. The DARKORANGE2 module had the strongest correlation with tumors. G Taking the intersection of the RED module, the DARKORANGE2 module, and the energy metabolism-related genes module, we found that PYGB, SCL2A1, and SLC16A3 are vital genes in PC. h–J The model consisting of these three genes was verified in the TCGA database to have a strong correlation with both diagnosis and prognosis of PC and is a risk factor for poor prognosis of PC. K Lasso coefficient profiles of the three PC prognostic genes. L Riskscore = (0.152) × PYGB + (0.0767) × SLC2A1 + (0.1482) × SLC16A3 for the three prognostic genes obtained using tenfold cross-validated lasso regression using minimum λ