Fig. 3

ROC and PR in the validation dataset. ROC and PR curves for model comparison across datasets. (A) ROC curves of six models on the validation dataset, demonstrating AMU’s superior discrimination ability (AUC = 0.953) compared to other methods. (B) Precision-Recall curves on the validation dataset showing AMU’s excellent performance (mAP = 0.972) in balancing precision and recall. (C) ROC curves on the testing dataset of post-treatment samples, where AMU maintains relatively robust performance (AUC = 0.672) despite the challenging transfer scenario. (D) PR curves on the testing dataset highlighting AMU’s ability to maintain precision (mAP = 0.800) even with treatment-modified expression profiles. The diagonal line in ROC curves represents random prediction (AUC = 0.5), while higher curves indicate better performance. For PR curves, higher curves represent superior precision-recall balance across threshold values.