Artificial intelligence at the gut-oral microbiota frontier: mapping machine learning tools for gastric cancer risk prediction.

backgroundGastric cancer (GC) remains a significant global health burden, with high mortality due to delayed diagnosis. Advances in microbiome profiling and artificial intelligence (AI) have opened new frontiers in non-invasive cancer risk prediction. However, the methodological landscape of AI-driven microbiome-based GC prediction remains fragmented and poorly standardized.

objectiveTo systematically review and critically evaluate artificial intelligence (AI) and machine learning (ML) models developed for gastric cancer prediction using microbial and non-invasive biomarkers, spanning gut, gastric mucosal, and oral ecosystems as well as tongue-based imaging proxies. We aimed to map methodological rigor, translational readiness, and biomarker convergence across these domains.

methodsWe systematically searched PubMed, Scopus, and Web of Science for peer-reviewed studies published up to March 2025. Eligible studies applied ML or deep learning models to microbiome datasets for GC diagnosis, risk classification, or treatment response. Data extraction included sample source, sequencing method, taxonomic resolution, ML model type, validation strategy, performance metrics, interpretability tools, and reported microbial taxa. Descriptive synthesis, thematic clustering, and readiness scoring were conducted using structured visual analytics.

resultsNine studies met the inclusion criteria. Sample sources included gastric mucosa, feces, saliva, tongue coating, and tumor tissue. 16S rRNA sequencing was most common, with models primarily trained on genus-level data. Random Forest was the most frequently used algorithm (44.4%), followed by LASSO, LightGBM, and deep learning. AUC values ranged from 0.88 to 0.97 in validated models. However, only 33.3% of studies employed external validation, and interpretability and reporting standards varied widely. A Clinical Readiness Matrix and Validation Quality Assessment highlighted key translational gaps. Recurrent microbial biomarkers included Veillonella, Fusobacterium, Prevotella, and Porphyromonas.

conclusionAI-based microbiome models, including non-invasive diagnostics, show high potential for gastric cancer prediction. Yet, reproducibility, external validation, and reporting transparency remain critical barriers to clinical implementation. Standardized pipelines, multi-omics integration, and prospective validation are needed to transition this field from proof-of-concept to precision oncology.