Classification model based on salivary microbiota profile

We performed a supervised classification using a cross-validated random forest model, based on the pre-operative OTUs differing between non-inflammation and inflammation samples [27–29]. The relative abundances of different OTUs were processed using z-score standardization (mean 0, standard deviation 1). Three-fold cross-validation was performed on the random forest model (Weka 3.7.12, http://www.cs.waikato.ac.nz/ml/weka/index.html) using the different OTU abundance profiles with a relative abundance higher than 0.01% to assign saliva subjects to two groups (non-inflammation and inflammation).