AI-driven discovery of novel extracellular matrix biomarkers in pelvic organ prolapse
Fig 4
This figure presents a detailed evaluation of EPOP’s performance on the test set, combining core classification metrics with a thorough analysis of threshold effects.
A: Precision-Recall curve (AUPRC = 0.999) demonstrating near-perfect performance with high precision maintained across most recall values. B: ROC curve (AUROC = 0.999) illustrating excellent classification performance, substantially outperforming random classification. C: Score distribution by class, showing distinct prediction score distributions for ECM and non-ECM proteins. D: Confusion matrix at the optimal classification threshold (0.168), providing a detailed breakdown of classification outcomes. E: Performance metrics (Accuracy, F1-score, Precision, Recall) across varying classification thresholds, highlighting the optimal threshold (0.168) that maximizes F1-score. F: Cumulative score distributions for ECM and non-ECM proteins, visually demonstrating the model’s high confidence and clear class separation.