Diversity and random forest models of oral microbiomes in periodontal health using publicly available data

BackgroundEvidence on the 16S metabarcoding of supragingival, subgingival, and salivary microbiomes in periodontal health remains limited. We aimed to analyze the diversity and potential of machine‐learning models of supragingival, subgingival, and salivary microbiomes in periodontal health.MethodsA total of 848 samples (supragingival = 210; subgingival = 155; saliva = 483) from 491 periodontally healthy subjects were included. Publicly available Illumina sequences were processed with mothur, and taxonomy was assigned using an oral‐specific database. Random forest (RF) models were built on the training set (2/3 of the samples) using a 3‐fold cross‐validation. They were tested on the test set (1/3).ResultsA total of 121 amplicon sequence variants (ASVs) presented with differential abundances between the two types of plaque, 212 between the supragingival and saliva samples, and 160 between the subgingival and saliva (p < 0.01). Furthermore, the supragingival versus subgingival model consisted of five ASVs. The performance parameters on the test set were area under the curve (AUC) = 0.908, accuracy (ACC) = 84.30%, sensitivity = 95.71%, and specificity = 68.63%. Both the supragingival and subgingival versus saliva models also had five ASVs. These two models revealed similar performance (AUC = 0.992 and 0.986, ACC > 95%, sensitivity > 90%, specificity > 95%).ConclusionAlthough supragingival and subgingival bacterial profiles diverged only modestly, primarily due to taxa with small effect sizes, they were both compositionally distinct from the salivary microbiome. RF models accurately classified samples by niche, with higher performance in distinguishing saliva from plaques. Specific ASVs from Escherichia, Fusobacterium, Granulicatella, Treponema, Peptostreptococcaceae [XI][G‐9], and Prevotella were identified in subgingival plaque, while Oribacterium and Solobacterium were identified in saliva, indicating potential niche‐specific microbial signatures in periodontal health.Plain Language SummaryMapping oral microbes in relation to periodontal health is essential for microbiome‐based diagnostics and the development of new preventive/therapeutic strategies. Our two‐by‐two predictive models demonstrated that a small set of bacterial ASVs can accurately classify periodontally healthy samples according to their oral niche. Notably, models distinguishing saliva from dental plaques achieved superior performance compared to those discriminating between plaques. This likely reflects the greater resemblance in dominant microbial taxa between the two plaque niches. These findings underscore the potential of machine‐learning approaches to identify key microbial signatures and highlight the predictive ASVs as promising biomarkers for characterizing oral niches in periodontal health.