Fig. 3

ROC curves and feature importances of ML models in predicting LLPS behavior. A The LR, KNN, SVM, GaussianNB, RF, LightGBM, and AdaBoost models were evaluated using repeated SG10CV, and their performances were assessed using ROC curves. The average curve for each SG10CV iteration of the repeated SG10CV was calculated, and the final result is shown as the total average curve. The values in brackets represent the overall average ROC-AUC values. B The top 10 average feature importances of the repeated SG10CV for the AdaBoost model are shown in the figure. The average value was calculated for each model trained within each SG10CV, and the average across all SG10CVs was calculated for the final result. Error bars represent standard deviations within each SG10CV