Fig. 5. Baseline fecal microbial abundances and SCFA and the risk of PTE.

(A) Receiver operating characteristics (ROC) curve for a random forest classifier predicting PTE from baseline abundances of 4 fecal microbes, selected from the 25 ASVs associated with epilepsy with q<0.25. Classifier performance was assessed by 5-fold cross-validation and is shown as the area under the ROC curve (AUC). The 95% confidence interval (shown as a blue area) was determined by bootstrapping. (B) Importance scores are shown for the four ASVs in the classifier shown in (A), representing their contribution to classifier accuracy. ASV color indicates whether the ASV was enriched (red) or depleted (green) prior to LFPI in rats that proceeded to develop PTE after LFPI. (C, D) ROC curves for PTE prevention derived from multiple logistic regression of 50 samples collected before LFPI. (C) When all 8 SCFAs are considered, AUC is 0.7222. (D) To avoid overfitting, we reduced the number of SCFA, and the best prediction (AUC=0.7167) was observed when we considered four SCFAs: 2-methylbutyrate (2Meth), isovaleric (Isov), isobutyric (Isobut), and valeric. Shown are ROCs for each individual SCFA and the 4 SCFAs combined (4 SCFA). (E) ROC curve for a random forest classifier predicting PTE from baseline abundances of four fecal microbes and four fecal SCFAs. (F) Importance scores are shown for the ASVs and SCFAs in the classifier shown in (E).