Artificial Intelligence Driven Salivary Multi-Omics Biomarkers for Early Detection of Periodontitis: A Systematic Review

Background

Periodontitis is a highly prevalent chronic disease, with severe forms affecting approximately 10% of adults worldwide. Conventional diagnostic methods, including periodontal probing and radiography, detect disease only after substantial tissue destruction. Saliva contains diverse biomarkers reflecting inflammation and tissue breakdown, while artificial intelligence (AI) can integrate multi-omics datasets to support earlier diagnosis. This systematic review synthesizes current evidence on AI-driven salivary multi-omics biomarkers for early detection of periodontitis.

Methods

We have searched PubMed, Scopus, Web of Science, Embase, and Cochrane (January 2020–2026) for peer-reviewed human studies. Eligible studies included systematic reviews, meta-analyses, randomized trials, prospective cohorts, diagnostic accuracy studies, and observational studies evaluating salivary biomarkers and AI for periodontitis. Two reviewers independently screened studies, extracted data, assessed methodological quality using validated tools, and performed a qualitative evidence synthesis.

Results

Frequently investigated biomarkers included IL-1β, TNF-α, MMP-8, MMP-9, CRP and oxidative stress markers. Proteomic studies consistently identified complement C3, S100A8, and fibrinogen, although diagnostic reporting remained limited (S100A8 pooled AUC≈0.71). Salivary microRNAs proved higher diagnostic performance, with pooled sensitivity of approximately 0.85–0.87, specificity of 0.83–0.88, and HSROC AUC of 0.90–0.93 for miR-146/miR-155. Metabolomic evidence remained limited. Salivary microbiome-based LightGBM models achieved AUCs of 0.81–0.87. Random forests and gradient boosting were the major AI approaches, while integrated multi-omics models generally outperformed single-marker models (pooled AUC≈0.94). Current evidence remains heterogeneous, with limited external validation, inconsistent protocols and challenges in reproducibility and interpretability.

Discussion

AI-assisted salivary diagnostics support precision periodontology, but clinical translation remains constrained by methodological heterogeneity, small cohorts, and limited multicenter validation. Standardized sampling, assay harmonization, and prospective validation of integrated biomarker–AI models are required before routine implementation.

Conclusion

AI-driven salivary multi-omics models demonstrate high diagnostic accuracy in research settings; however, heterogeneous evidence and limited clinical validation currently preclude routine clinical implementation.