Fig. 5.

Construction and validation of the purity-independent regulatory network. A Heatmap displaying eligible lncRNA–miRNA–mRNA pairs supported by GDCRNAtools. Only experiments confirming RNA relationships were included. RNA relationships with consistent evidence support across datasets with varying tumor purity levels were considered preserved in pancreatic cancer (n = 3). B Model coefficients (top axis) versus the logarithm of lambda (bottom axis) for LASSO analysis of the construction group (n = 540). The coefficients of three genes were non-zero at lambda.1se, which represents the lambda value generating the most regularized model with cross-validated errors within one standard error of the minimum. C Three-fold cross-validation C-indexes versus the logarithm of lambda (bottom axis) and the number of non-zero coefficients (top axis) in the construction group (n = 540). D Kaplan–Meier curve illustrating the high- and low-risk groups in the construction model using the three-gene model (n = 540, log-rank test, P < 0.0001). E Kaplan–Meier curve illustrating the high- and low-risk groups in the validation model using the three-gene model (n = 288, log-rank test, P = 0.0067). F Risk score (RS) panel of the construction model (n = 540). G Risk score (RS) panel of the validation model (n = 288)