Fig. 2

Results of Step One and Step Two model development. (a) PCA plot of 119 samples with 865,859 CpG sites. (b) PCA plot of 48 samples with 866,238 CpG sites. (c) The value of AUPRC in 2 to 219 feature number of PCA model. The dashed line reveals the 30 features corresponding 0.95 AUPRC. (d) PCA of training data with 30 features (CpG sites). The dashed line indicates the classification threshold for normal and tumor. (e) The precision-recall curve in the PCA model. The red dot indicates the threshold corresponding to precision and recall. (f) Confusion matrix of prediction results in the PCA model of training data. (g) Confusion matrix of prediction results in the gradient boosting model by using training data. (h) In the precision-recall curve of the gradient boosting model, the red dot indicates the threshold corresponding to precision and recall