Fig. 2

Predictive models for detection metastasis of ESCC by 5hmC biomarkers in cfDNA. A Schematic of the machine learning process. B Heatmaps illustrating the distribution of the 14 5hmC markers in training cohorts across patients exhibiting either metastasis or non-metastasis, with demographic data such as age and sex included. C Heatmaps illustrating the distribution of the 14 5hmC markers in validation cohorts across patients exhibiting either metastasis or non-metastasis, with demographic data such as age and sex included. D ROC curves of the metastasis classification system applying the 14 5hmC markers in training cohorts, depicting the true positive rate (sensitivity) against the false positive rate (1-specificity). E ROC curves of the metastasis classification system applying the 14 5hmC markers in validation cohorts, depicting the true positive rate (sensitivity) against the false positive rate (1-specificity). (F-G) Confusion matrices from the metastasis prediction model for each cohort