Fig. 4

Performance comparison of different machine learning regression models. In these two histograms (A, B), the results were obtained based on five-fold cross validation (A) and independent test (B) for the three targets. The R² and RMSE scores were used to evaluate the performance of different machine learning models including DNN, KNN, PLS, SVM RF and MT-DNN. In the scatter plots (C–E), each point stands for one molecule with its real pX (x-axis) and the predicted pX (y-axis) by the RF model which was chosen as the final predictors for A₁AR (C), A_2AAR (D) and hERG (E) based on five-fold cross validation (blue) and independent test (orange)