Gastric cancer heterogeneity profoundly impacts clinical outcomes, yet the molecular underpinnings remain incompletely characterized. Here we present a comprehensive single-cell atlas of gastric cancer, integrating five cohorts comprising 101 samples and 327,471 cells. Unsupervised clustering delineated seven transcriptionally distinct tumor cell populations. Among these, an ITGB1-expressing subpopulation displayed hallmarks of epithelial-mesenchymal transition and demonstrated robust associations with metastatic progression and adverse prognosis. Leveraging the ITGB1 + transcriptional signature, we stratified patients into three molecular classes: C1 (metabolism-dominant), C2 (proliferation-dominant), and C3 (invasion-dominant). C3 tumors exhibited pronounced metastatic capacity, stemness features, and immunosuppressive microenvironmental remodeling, with five-year survival of merely 23%. Mechanistic interrogation identified INMT (Indolethylamine N-Methyltransferase) as a central driver of the aggressive C3 phenotype. Genetic ablation of INMT attenuated proliferative and invasive behaviors in vitro and suppressed tumor growth in vivo. Structure-based virtual screening nominated Savolitinib as a candidate INMT-targeting agent, a prediction validated through molecular dynamics simulations demonstrating stable binding interactions. Pharmacological INMT inhibition recapitulated the phenotypes observed with genetic knockdown, reducing tumor burden in xenograft models. Our work establishes a single-cell-resolved molecular taxonomy for gastric cancer, elucidates the mechanistic connection between epithelial-mesenchymal transition programming and metastatic dissemination, and positions INMT as an actionable therapeutic vulnerability—collectively advancing the framework for precision oncology in this disease.