An aging-informed multi-layer bioinformatics framework prioritizes GKN1 and PLAU/uPA-associated candidates in gastric cancer

Gastric cancer is molecularly heterogeneous. We used curated aging-gene annotation as a candidate-filtering prior, rather than as evidence of an aging mechanism, within an integrated bulk and single-cell bioinformatics framework. Differentially expressed genes in GSE29272 were intersected with 542 genes curated in aging databases. LASSO and random forest prioritized hub genes, and an 11-gene logistic model was evaluated in GSE54129 using fixed training-set standardization parameters. For GSE163558, raw counts were aggregated by donor and by donor x cell type, normalized as TMM log2 counts per million (CPM), and compared directionally across PT1–PT3 versus NT1; formal group-level P values were not calculated because there was only one non-tumor donor. Epithelial-state ordering and pooled CellChat inference were used descriptively. DGIdb prioritization, molecular docking, MD trajectory analysis of two uPA inhibitors, and a PLAU-centered candidate RNA interaction network extended the therapeutic and regulatory assessment. Among 312 differentially expressed genes, 18 overlapped the curated aging-gene set, and GKN1 ranked first by random forest. In GSE54129, the 11-gene panel achieved an AUC of 0.979 (95% CI 0.960–0.999), accuracy of 0.803, sensitivity of 0.766, specificity of 1.000, and a Brier score of 0.174. In the donor-aggregated single-cell analysis, GKN1 was lower and PLAU was higher in each tumor donor relative to NT1; the mean tumor-minus-NT1 differences were − 11.41 and + 4.48 log2 CPM, respectively. The drug network contained 45 records, including 42 for PLAU. Upamostat and WX-UK1 had docking scores of − 8.1 and − 7.1 kcal/mol, respectively, and showed complementary MD trajectory profiles. The PLAU-centered candidate RNA interaction network comprised 24 nodes and 41 database-supported edges. The integrated analysis prioritizes GKN1 as a candidate marker of gastric epithelial-state loss and PLAU/uPA as a drug-associated tumor-microenvironment axis. These findings generate testable hypotheses but do not establish an aging mechanism, causal intercellular signaling, or clinical utility; donor-balanced and experimental validation is required.