BACKGROUND: Early detection of gastric cancer (GC) and reliable risk stratification for metachronous gastric lesions (MGLs) remain societal and clinical challenges, particularly in intermediate risk populations. Non-invasive approaches such as saliva-based biomarkers could complement current strategies. The aim of this study was to identify and validate a tissue-based gene expression signature for early gastric lesions, explore its potential for MGL prediction, and assess its detectability in saliva. METHODS: Three studies were conducted: (1) a retrospective case control study to identify (RNA sequencing with machine learning) and validate (reverse transcription [RT] quantitative polymerase chain reaction [qPCR]) a gene expression signature for early gastric cancer using formalin-fixed paraffin-embedded (FFPE) samples, (2) a retrospective longitudinal study evaluating the ability of the signature to stratify MGL risk, and (3) a prospective study testing the signature in saliva using droplet digital (dd)PCR in patients with gastric lesions and endoscopy-confirmed controls. RESULTS: A six-gene tissue-based expression signature (ADAMTSL1, CCNA2, HSP90AB1, HSPD1, PSAPL1, and SLC5A5) robustly discriminated early gastric lesions from non-tumor mucosa (area under the curve (AUC) = 0.96 and 95% confidence interval [CI]: 0.94-0.99). Models tailored for MGL prediction, incorporating clinical variables, achieved moderate performance (AUC = 0.74 and 95% CI: 0.59-0.88). In saliva, only the SLC5A5 gene showed consistent dysregulation. When combined with age and sex, the model reached an AUC of 0.78 (95% CI: 0.69-0.88) for the non-invasive detection of early GC, with a positive predictive value of 0.69 and negative predictive value of 0.81. CONCLUSION: This study presents a validated tissue-based gene signature for early GC detection and exploratory MGL risk stratification. Salivary SLC5A5 shows potential as a non-invasive biomarker, though its utility requires further validation in dedicated saliva-based studies.