Prediction model for periodontitis stage based on the salivary microbiome

Abstract

This study aimed to characterize salivary microbiome compositions that can classify periodontal health and various stages of periodontitis. We collected saliva samples from 250 study subjects, including 100 periodontally healthy controls and 150 periodontitis patients in stages I/II/III. We performed 16S ribosomal RNA gene sequencing to characterize their salivary microbiomes. Alpha diversities show significant differences between healthy and periodontitis. Differentially abundant taxa were identified by ANCOM. Random forest machine learning models were used to classify each periodontitis stage based on the centered log-ratio of differentially abundant taxa. We identified 20 differentially abundant taxa among the groups in the salivary microbiomes of all groups. Among these differentially abundant taxa, Porphyromonas gingivalis and Actinomyces spp. are the most important taxa on the random forest model to classify the periodontitis statuses. Our random forest model classified multiple periodontitis statuses with an area-under-curve of 0.829 ± 0.124, sensitivity 0.884 ± 0.022, and specificity 0.652 ± 0.065. Moreover, because it can be difficult to diagnose in dentistry practice, we performed our classifier model to distinguish healthy or stage I, providing an area-under-curve of 0.736 ± 0.168, sensitivity 0.789 ± 0.102, and specificity 0.622 ± 0.196. Furthermore, our random forest model detected periodontitis patients from healthy individuals with an area-under-curve of 0.924 ± 0.088, sensitivity of 0.862 ± 0.175, and specificity of 0.921 ± 0.061. Finally, we evaluated our classification model with external data sets from Spanish and Portuguese subjects. Some evaluations showed a slight decrease, but it might be due to different salivary microbiome compositions from ethnicity. Significant differences were identified in the differentially abundant taxa among healthy controls and the various stages of periodontitis.IMPORTANCEPeriodontitis is a common but complex oral disease that can lead to tooth loss and contribute to systemic health issues. Early and accurate diagnosis is essential for effective intervention, yet traditional diagnostic methods often rely on invasive clinical assessments that may miss early signs. This study demonstrates that salivary microbiome profiles can be used to classify both periodontal health and multiple periodontitis stages using a machine learning approach. By identifying the 20 key microbial taxa, including Actinomyces spp., we developed a non-invasive predictive model with high diagnostic accuracy. Importantly, the model was also able to detect early-stage disease and performed well across external data sets, highlighting its potential for broader clinical application. These findings suggest that a salivary microbiome-based diagnostic tool may support more precise, accessible, and early diagnosis of periodontitis in dental disease management.