Figure 5.
Metabolite-level classification model for REP and NEP. (A) Volcano plot showing differential metabolite expression between REP and NEP groups. Red and blue points indicate significantly up- or downregulated metabolites, respectively, based on log2 fold change and -Log10 (p-value). (B) Venn diagram highlighting the overlap between significantly altered metabolites (t-test, p < 0.05, fold change > 1.5 or < 2/3) and those with VIP > 1 from PLS-DA. (C) Box plots of six key metabolites with statistical significance (*p < 0.05, **p < 0.01) between groups. (D) Random forest feature importance ranking of selected metabolites. (E) ROC curve (AUC = 0.875) showing model performance in classifying REP vs. NEP. (F) PR curve (AUC = 0.943) illustrating precision-recall characteristics. (G, H) Permutation test results for AUC and accuracy (1,000 iterations), with empirical p-values indicating statistical robustness. CSF, cerebrospinal fluid; NEP, non-epilepsy; REP, refractory epilepsy; PLS-DA, partial least-squares regression discriminant analysis; ROC, receiver operating characteristic; PR, Precision-recall; AUC, area under the ROC curve.