The potential value of oral microbial signatures for prediction of oral squamous cell carcinoma based on machine learning algorithms.

Abstract

This study aimed to explore the potential predictive value of oral microbial signatures for oral squamous cell carcinoma (OSCC) risk based on machine learning algorithms. The oral microbiome signatures were assessed in the unstimulated saliva samples of 80 OSCC patients and 179 healthy individuals using 16S rRNA gene sequencing. Four different machine learning classifiers were used to develop prediction models. Compared with control participants, OSCC patients had a higher microbial dysbiosis index (MDI, p < 0.001). Among four machine learning classifiers, random forest (RF) provided the best predictive performance, followed by the support vector machines, artificial neural networks and naive Bayes. After controlling the potential confounders using propensity score matching, the optimal RF model was further developed incorporating a minimal set of 20 bacteria genera, exhibiting better predictive performance than the MDI (AUC: 0.992 vs. 0.775, p < 0.001). The novel MDI and RF model developed in this study based on oral microbiome signatures may serve as noninvasive tools for predicting OSCC risk.