Extended Data Fig. 8. Receiver operating characteristic (ROC) curves for classifiers designed to separate RVI time points from healthy baselines at different discrimination thresholds.

a, b, Performance in the training (top) and test sets (bottom) for LR (a) and SVM (b) models. Plots show the true positive rate (y axis) versus the false positive rate (x axis) for each ’ome and all ’omes combined. AUC scores for ROC curves are listed on the right for classification accuracy. For binary classification between healthy and RVI data, combined multi-omes had the highest prediction performance, followed by the metabolome, compared to others in the test cohort. c, ROC curves and AUC scores based on all-pairs testing using the LR model for classifying RVI events. The plot shows the true positive rate (y axis) versus the false positive rate (x axis) for each pairwise combination. Bottom right, summary of all pairwise combinations of multi-omes and their respective AUC scores in percentages.