Skip to main content
. 2025 Sep 11;16:1630062. doi: 10.3389/fmicb.2025.1630062

Figure 6.

Panel A shows a heatmap of Spearman correlation values for various metabolites and bacterial genera. Panel B is a bar chart depicting feature importance with associated mean decrease accuracy values. Panel C displays an ROC curve with an AUC of 0.953. Panel D presents the distribution of AUC values from a permutation test. Panel E shows the accuracy distribution from a permutation test with an accuracy of 0.875. Panel F features a PR curve with an AUC of 0.976.

Correlation analysis of CSF metabolites with gut microbiota and classification model performance. (A) Spearman correlation heatmap between four CSF metabolites and 10 gut bacterial genera. Red and blue shading indicate positive and negative correlations, respectively, with asterisks marking statistical significance (*p < 0.05; **p < 0.01). (B) Feature importance plot ranking bacterial genera and metabolites based on mean decrease in accuracy from the random forest model. (C) ROC curve (AUC = 0.953) illustrating model performance in distinguishing REP from NEP. (D, E) Permutation tests for AUC and accuracy (1,000 iterations), with empirical p-values indicating statistical robustness. (F) PR curve (AUC = 0.976) showing precision-recall characteristics. CSF, cerebrospinal fluid; NEP, non-epilepsy; REP, refractory epilepsy; ROC, receiver operating characteristic; PR, Precision-recall; AUC, area under the curve.