Fig. 2: Performance of cancer detection.
From: Integrative modeling of tumor genomes and epigenomes for enhanced cancer diagnosis by cell-free DNA
A ROC curve for different models on the MGI training cohorts. A total of 1396 cancer and 417 normal control samples were used for training, and 59 cancer and 96 normal control samples were used for validation as the MGI validation cohort. B ROC curve for different models on the Illumina training cohort. A total of 573 cancer and 670 normal control samples were used for training, and 97 cancer and 58 normal control samples were used for validation as the Illumina validation cohort. C, D Sensitivity values with the 95% confidence interval at 95% specificity broken down by the tumor stage (left) and cancer type (right) for the C MGI training cohort and D Illumina training cohort. Confidence interval for sensitivity value was calculated from 1000 bootstraping samplings. A–D Our genome, epigenome, and combined models were compared with predictions based on fragmentation patterns9 (fragpattern), fragment size profiles8 (fragsize), and copy number variations3 (cnv). NA Stage information not available, CI Confidence interval. Source data are provided as a Source Data file.
