ObjectiveThis study aimed to establish a prognostic model related to prostate cancer (PCa) recurrence-free survival (RFS) and identify biomarkers.MethodsThe RFS prognostic model and key genes associated with PCa were established using Least Absolute Shrinkage and Selection Operator (LASSO) and Cox regression from the Cancer Genome Atlas (TCGA)-PRAD and the Gene Expression Omnibus (GEO) GSE46602 datasets. The weighted gene co-expression network (WGCNA) was used to analyze the obtained key modules and genes, and gene set enrichment analysis (GSEA) was performed. The phenotype and mechanism were verified in vitro.ResultsA total of 18 genes were obtained by LASSO regression, and an RFS model was established and verified (TCGA, AUC: 0.774; GSE70768, AUC: 0.759). Three key genes were obtained using multivariate Cox regression. WGCNA analysis obtained the blue module closely related to the Gleason score (cor = –0.22, P = 3.3e − 05) and the unique gene glutathione peroxidase 2 (GPX2). Immunohistochemical analysis showed that the expression of GPX2 was significantly higher in patients with PCa than in patients with benign prostatic hyperplasia (P < 0.05), but there was no significant correlation with the Gleason score (GSE46602 and GSE6919 verified), which was also verified in the GSE46602 and GSE6919 datasets. The GSEA results showed that GPX2 expression was mainly related to the epithelial–mesenchymal transition (EMT) and Wnt pathways. Additionally, GPX2 expression significantly correlated with eight kinds of immune cells. In human PCa cell lines LNCaP and 22RV1, si-GPX2 inhibited proliferation and invasion, and induced apoptosis when compared with si-NC. The protein expression of Wnt3a, glycogen synthase kinase 3β (GSK3β), phosphorylated (p)-GSK3β, β-catenin, p-β-catenin, c-myc, cyclin D1, and vimentin decreased; the expression of E-cadherin increased; and the results for over-GPX2 were opposite to those for over-NC. The protein expression of GPX2 decreased, and β-catenin was unchanged in the si-GPX2+ SKL2001 group compared with the si-NC group.ConclusionWe successfully constructed the PCa RFS prognostic model, obtained RFS-related biomarker GPX2, and found that GPX2 regulated PCa progression and triggered Wnt/β-catenin/EMT pathway molecular changes.