Fig. 2

cfDNA-FEP performance. A ROC curve for the training set generated with 10-times 10-fold cross-validation (AUC = 0.91). Mean (solid line) and range (shaded area) are plotted. The dashed line denotes the theoretical performance of a random classifier. B Cancer scores predicted by the cfDNA-FEP classifier on the test set for COAD, RCC, and healthy samples stratified by clinical stage. A decision cutoff of 0.5 is denoted as a dashed line. C ROC curve for the testing set (AUC = 0.94). The dashed line denotes the theoretical performance of a random classifier. D The heatmap of fragmentomic features and characteristics of the test set samples (n = 52). cfDNA-FEP classification results are shown as cancer score and predicted class (cancer or health)