Revolutionizing salivary biomarkers through machine learning and artificial intelligence

Chronic diseases remain major global health burdens, and early detection is essential for preventing progression and reducing complications. Saliva, a non-invasive, information-rich, and easily preserved biofluid, offers a viable alternative to invasive and costly traditional screening methods such as cytopathology and tissue biopsy. Recent advances in multi-omics technologies-including proteomics, transcriptomics, genomics, metabolomics, microbiomics, and exosome profiling-have uncovered diverse salivary biomarkers linked to oral and systemic diseases. However, the high dimensionality and complexity of these datasets require robust computational methods to extract clinically meaningful patterns. Machine learning (ML), as a key component of artificial intelligence (AI), provides such analytical capability by integrating heterogeneous data and improving biomarker-based classification performance. This review summarizes studies from 2010 to 2025 on AI-enabled saliva diagnostics, outlining major biomarker categories-including proteins, nucleic acids, metabolites, exosomes, and microbial signatures and commonly used algorithms. Key performance metrics such as accuracy (ACC), area under the receiver operating characteristic curve (AUC), sensitivity (SE), and specificity (SP) are discussed, together with methodological limitations. Current evidence supports the strong potential of ML-assisted salivary biomarker analysis to enhance early disease screening. These advances may help establish saliva as an accessible tool for large-scale prevention, particularly in resource-limited or high-risk settings.