Fig. 8

Receiver operating characteristic (ROC) curves (a–d), showing the ability of the potential biomarkers to distinguish between control subjects and FH patients, as well as heatmaps data visualization to depict variance across multiple variables (e, f). a, b Multivariate ROC curves of the six models in primary and validation cohorts have been generated employing random forest prediction model with combined features. We found that the model with 10 features entered showed excellent predictive performance with ROC AUC (area under the curve) values > 99% for both discovery and validation data. c, d 15 significant features were ranked based on their average importance (in group classification) during cross validation. e, f Color-coded maps that illustrate the relative abundance of feature combinations for two independent cohorts of cases