Early non-invasive detection is crucial for improving the prognosis of gastric cancer (GC). Dysbiosis in the oral and gastric microbiome is closely associated with GC development, yet its complex nature poses challenges for traditional analytical methods. This study aims to integrate oral and gastric microbial characteristics and employ machine learning algorithms to construct a high-precision GC diagnostic model. We collected saliva, gastric fluid, and gastric mucosa samples from 106 GC patients and 111 healthy controls. Microbiome data were obtained via 16S rRNA sequencing, with analyses conducted on species abundance, diversity and community composition. Algorithms including random forest and support vector machines were employed to identify the most discriminative bacterial genera. Multiple diagnostic models were trained based on these findings and evaluated through ten-fold cross-validation and independent external datasets. Results revealed significant differences in microbial composition between GC patients and healthy individuals. Diagnostic models constructed based on key bacterial genera demonstrated excellent performance: AUC values in the training set reached 0.96, 0.87, and 0.96 for oral, gastric fluid and gastric mucosa classifiers respectively, maintaining robust performance in external validation. Functional prediction revealed dysregulation in the microbiota of GC patients, while correlation analysis further indicated symbiotic relationships among cancer-associated bacterial genera. This study successfully established a GC diagnostic model based on oral and gastric microbial signatures. Its outstanding performance validates the substantial potential of the ‘oral-stomach’ microbial axis in GC auxiliary diagnosis, offering a novel cost-effective strategy for early screening.
Lei et al. (2026) studied this question.