Figure 1

Patient cohort and sample processing. (A) Patient clinical characteristics. (B) Included faeces and saliva samples were investigated regarding their bacterial and microbiome-derived metabolites composition. Taxonomy was obtained based on 16S rRNA sequencing and Naive Bayes classification algorithm. Microbiome-derived metabolites were inferred by community Flux Balance Analysis (FBA) and metagenomic gene content was inferred using the PICRUSt2 algorithm28. Oral metabolites profiles were used to construct a classification model that was further validated using an independent cohort.