Fig. 3

The random forest model constructed using 12 genera can be used as a disease classifier to differentiate JIA patients from healthy controls. a Plot of genera numbers vs error rates. As the genera numbers increased, the error rates decreased sharply. The dashed gray line marks the optimal cut-off for biomarker selection. This analysis indicated that 12 was the optimal predictor (genus) number. b The variable importance of the genera analyzed using the randomForest package in R. The most important 12 genera are listed in the plot. The greater the Gini indices, the more important the variables are. c The relative abundance of the 12 genera identified by the random forest model and Wilcoxon test. The 4 genera marked with an asterisk differed significantly in abundance between the two groups by Wilcoxon test (corrected P < 0.05). d ROC of the random forest model constructed using the 12 genera. The diagonal line in the graph marks an AUC of 0.5. The 95% confidence intervals are shown as shaded areas. e DCA for the random forest model constructed using the 12 genera. The y-axis measures the net benefit. The green line represents the situation with the assumption that all children received treatment due to JIA. The blue line indicates the net benefit under the assumption that no children received treatment due to JIA (e.g., representing the natural disease course without medical intervention so that the net benefit is constantly zero). The red line is above the green and blue lines, especially within the threshold probability of 0.23–0.77, which implies that the prediction model is able to achieve a greater net benefit than the situation when the children are treated or untreated without any model