Performance of salivary microbiota in detecting periodontitis using a machine learning approach.

Abstract

Altered salivary microbiota due to the progression of periodontitis may serve as a marker for simple and accurate identification of periodontitis. In this study, we examined saliva samples collected from 2,050 community-dwelling adults using 16S rRNA gene sequencing and verified the predictive performance of salivary microbiota in detecting periodontitis using a light gradient boosting machine algorithm. Five-fold stratified cross-validation was applied with 10 iterations, and the predictive performance was evaluated using the mean area under the receiver operating characteristic curve (AUC) value. In detecting periodontitis defined by number of teeth with probing depth ≥4 mm, localized (≥2 teeth), intermediate (≥4 teeth), and generalized (≥6 teeth) cases were detected with mean AUC values of 0.81 (95% confidence intervals, 0.80-0.81), 0.85 (0.84-0.86), and 0.87 (0.87-0.88), showing an increasing trend with extent. According to the Shapley additive explanation analysis, Porphromonas gingivalis, Tannerella forsythia, Mycoplasma faucium, Treponema species HMT-237, and Fretibacterium species HMT-362 were identified as important features for the detection of periodontitis. Our study presents the potential of salivary microbiota as a tool for mass screening of periodontitis and provides information on novel and important targets, including taxa other than known periodontal pathogens, to establish salivary screening tests.