Fig. 1
From: Analysis of the benefits of imputation models over traditional QSAR models for toxicity prediction
Comparison of data splitting strategies. For the compound-based split, 20% of the compounds were randomly selected for the test set. These compounds were used for testing across all the assays. For the assay-based split, each assay was considered independently and 20% of the compounds were selected at random and placed in the test set. This resulted in a different set of test compounds for each assay. A assay, C compound, 0 non-toxic, 1 toxic