Fig. 2

Marker profiles and model performance. A Feature profiles in the training and validation cohorts. The heatmaps display the selected biomarkers, including the mutation status of 7 genes, 23 DNA methylation markers, 14 5’ end motifs, and CNVs in 34 chromosome arms. The data of DNA methylation, end motif and CNV in the heatmaps were centered and scaled in the row direction. B Receiver operating characteristic (ROC) curves for the integrated model (integrating DNA methylation, 5’ end motif, CNV and gene mutation), each individual model (DNA methylation, 5’ end motif, CNV, or gene mutation) and plasma CEA assay in the training (left) and validation (right) cohorts. The plasma samples of 93 CRC and 73 healthy participants in the training cohort, and 89 CRC and 74 healthy participants in the validation cohort were available for the plasma CEA assay. C The predicted probabilities for healthy individuals and stage I-IV CRC patients predicted by the integrated model (integrating DNA methylation, 5’ end motif, CNV and gene mutation) and each individual model (DNA methylation, 5’ end motif or CNV) in the validation cohort. Participants with predicted probabilities close to 1 are more likely to have CRC based on the models. D The predicted results, presented as the probabilities predicted by each model minus the corresponding cutoff values (0.63 for DNA methylation model, 0.61 for end motif model, 0.81 for CNV model, and 0.51 for the integrated model), in the validation cohort. The reddish signals indicate the sample is predicted to be CRC positive based on the corresponding model, while the bluish signals indicate the sample is predicted to be normal/CRC negative based on the corresponding model. E The positive rates for plasma CEA and CA19-9 assays and our integrated CRC screening model in healthy individuals and stage I-IV CRC patients in the validation cohort. The plasma samples of 89 CRC and 74 healthy participants in the validation cohort were available for the plasma protein assay