Fig. 2

Taxa-wide supervised 5-repeat, 10-fold random forest classification models predict future early childhood caries status when using 12- (n = 158) and 24-month (n = 133) 16S rRNA gene amplicon sequenced saliva samples of children from Appalachia in an incidence density sampled case-control study (Center for Oral Health Research in Appalachia 2 cohort). A Area under the curve receiver operating curves from supervised random forests predicting future early childhood caries using the 273 most prevalent and abundant amplicon sequence variants at 12 months (black line) and 24 months (grey line). B Importance plots showing the top ten most important amplicon sequence variants from the 12- and 24-month supervised random forest classifiers performed on 273 amplicon sequence variants, as determined by mean decrease in the Gini coefficient, with the importance of the S. mutans amplicon included for comparison. C Joy plots showing the relative abundance distribution of the top 10 most important amplicon sequence variants and S. mutans among cases (black) and controls (grey) at the 12- and 24-month visits