Fig. 3

Performance of the HFmeRisk model. a AUC results of the prediction performance according to different features in the testing set. “(HFmeRisk/EHR/CpG model)” indicates the model with EHR and DNA methylation data, the model with DNA methylation data only, and the model with EHR data only, respectively. b Calibration plot of the DeepFM model in the testing set using 30 features. The Hosmer–Lemeshow statistic was 6.17, with P = 0.632. c Decision curve analyses of the HFmeRisk, 5 EHR model risk and 25 CpGs model risk in the testing cohort. d AUC results for the HFmeRisk model versus the Willliam’s model in male/female participants. e The association of CpG (cg10083824/cg03233656) and its DMG expression (GRM4/SLC1A4) in blood samples of FHS participants. X-axis is beta value of DNA methylation, Y-axis is expression value of RNA data. Rug plots display individual cases in X- and Y-axis. The smooth curve shows linear smooths in case/control status. The Pearson's correlation between CpG and DMG is driven mainly by case–control status. DMG, differentially methylated gene. The triangle represents the no-CHF participants; the dot represents the HFpEF participants