Fig 1: The development of the machine learning diagnostic models and key OSRGs. (A, B) Among the 31 DE-OSRGs, ACTA2, CYP17A1, ENO2, NGF, NR3C1, SELP, and SPP1 were identified with a connectivity score greater than five by least absolute selection and shrinkage operator (LASSO) algorithm. (C, D) CYP17A1, NR3C1, ENO2, NGF, and CCL2 were pinpointed via the Support vector machine recursive feature elimination (SVM-RFE) algorithm. (E) CYP17A1, NR3C1, ENO2, and NGF were obtained by overlapping both machine learning results as key OSRGs. (F) Receiver operating characteristic (ROC) curve of the four key OSRGs. (G) Expression levels of four key OSRGs were compared in the combined dataset. *P < 0.05, ***P < 0.001, ****P < 0.0001.