Fig. 3: Performance of HRDCNA model.
a, b ROC curves a and PR curves b analysis showed the performance of HRDCNA built using CNA features on the training and held-out datasets. The gray shaded area represents a 95% confidence interval. c HRDCNA scores for 1470 pan-cancer samples in training and held-out datasets are ordered from lowest to highest. The HRDCNA scores of almost all HRD samples are above 0.2. d, e The ROC curve d and PR curve e showed the performance of HRDCNA model in the independent validation dataset of 633 cancer samples. The gray shaded area represents a 95% confidence interval. f HRDCNA scores for 633 pan-cancer samples in three independent validation datasets are ordered from lowest to highest. The horizontal dashed line shows a cut-off score of 0.2. g The performance of HRDCNA model in training and held-out datasets is visualized with a confusion matrix under cut-off score of 0.2. h Kaplan-Meier (KM) survival analysis. Patients for analysis included 85 patients who had received platinum chemotherapy from 501 pan-cancer cohort. Samples are grouped based on HRDCNA scores. Patients with high HRDCNA scores showed significantly improved survival when treated with platinum chemotherapy. P value is calculated using log-rank test.
