Fig. 4
From: Analysis of the benefits of imputation models over traditional QSAR models for toxicity prediction
Performance of FN models. A Ames. B Tox21. The plots compare the MCC scores of XGB-FN models to single task XGB models. The first FN model in each plot corresponds to the situation where predicted values are used for the labels of the auxiliary assays of test set compounds, whereas in the second case experimental labels are used in place of the predicted values for the auxiliary assays, when these are available. Each box summarizes the MCC scores of 20 independent runs of the model on the test set with identical hyperparameters but different random seeds.