Figure 4 from Comparative Analysis of Cell-Free DNA Fragmentation Patterns in Canines with Sarcoma and Tumor-Free Canines and Humans

Cancer sample classification performance. ROC plots for the classification of healthy and cancer-derived plasma samples using individual features including short index by qPCR (blue), short fragment proportion by low-pass whole-genome sequencing (LPWGS; orange), long fragment proportion by LPWGS (gray), tumor fraction by ichorCNA (purple), and a random forest combining all features (black). The model was trained using plasma samples from healthy dogs and pretreatment samples from dogs with sarcomas. A, Hold-one-dog-out cross-validation performance on all samples, including posttreatment samples. B, Confusion matrices at specificities of 0.95 (left) and 0.90 (right) for all samples analyzed, showing the number of correctly classified samples in green and incorrectly classified samples in gray. C, Classification performance using hold-one-out cross-validation on pretreatment samples only. D, Confusion matrices at specificities of 0.95 (left) and 0.90 (right) for the cross-validation analysis of pretreatment samples only.