Fig. 1: The clinical prediction using the PCaseek-D and PCaseek-G models.
a, b Unsupervised hierarchical clustering of the proportion of the top 10% of PCa-specific hyper (a)/hypo (b) pDMRs, ranked by differential methylation values, between PCa tissues (n = 25) and adjacent normal tissues (n = 25). c, d Scatterplots of AUC values for the top 100 PCa-specific hyper (c) and hypo (d) pDMRs and mDMRs, with each point representing the classification performance of a pDMR or mDMR region between PCa tissues (n = 25) and adjacent normal tissues (n = 25). e Radar chart of the relationship between the threshold for the number of selected hypomethylated PCa-specific pDMRs and the accuracy of the PCaseek-D model on the validation cohort (ntumor = 47, nbenign = 44). f–h Scatterplots of the distribution of PCaseek-D scores for the control group and tumor group (f) with different PSA levels (g) and with low GS (Gleason score ≤ 3 + 4) and high GS (Gleason score ≥ 4 + 3) (h). The control group consists of healthy individuals and benign prostatic disease patients. i, j ROC curves of the PCaseek-D model on the training cohort (i) and the validation cohort (j). k Correlation matrix plot of PCaseek-D scores at various sequencing depths on the validation cohort. l The ROC curve and corresponding AUC values of PCaseek-D, tPSA, %fPSA, and PSAD on the validation cohort. m, n Bar charts of the specificity (m) and accuracy (n) of PCaseek-D, tPSA, %fPSA, and PSAD on the validation cohort. o The ROC curves and corresponding AUC values of PCaseek-D, tPSA, and %fPSA for patients with gray zone PSA. p The ROC curve and corresponding AUC value of PCaseek-G for the classification of low GS and high GS cases in the validation cohort. For c, d, f‒h, n.s. nonsignificant, *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 by two-sided Wilcoxon tests.
