Fig. 3

The deep learning features selection. A The distribution of the lowest mean squared error (MSE) with the corresponding penalization lambda value in LASSO-penalized model. B LASSO coefficient profiles of all deep learning features and the selected criterion of lambda. C Feature weight bar chart for the LASSO-penalized model. D The AUCs distribution of seven machine learning classifiers (SVM, KNN, Decision-Tree, Random-Forest, Extra-Tree, XGBoost, and LightGBM) with tenfold cross-validation. E The accuracy distribution of seven machine learning classifiers in train and test datasets. F The ROC curves of 7 machine learnings in TCGA-UVM cohort. G The ROC curves of 7 machine learnings in HX cohort. H The confusion matrix of SVM classifier in TCGA-UVM cohort