Fig. 3.

Various viral infections are associated with specific signatures of microbiome disruption. A The random forest model was used to predict the participants infected with FluA (n = 22), FluB (n = 138), RSV (n = 34), HRV (n = 61), or any virus (n = 349) from healthy (n = 98), and the ROC was calculated. B All samples of the five types were selected respectively; the selected samples and healthy samples were used for random (n = 100) under-sampling to fit a random forest model, and the AUC was calculated to evaluate the prediction performance. C The Top10-taxa with a higher value of Mean Decrease Accuracy from each categorical group was used for analysis