Fig. 4

Feature selection with permutation feature importance. A The area under the curve (AUC) calculated for a varying number of features selected by the permutation feature importance algorithm against the PubChem BioAssay dataset. B The composition of feature vectors changing with the vector length. Feature vectors comprise four groups of features calculated for drugs with Mol2vec (green), proteins with ProtVec (purple), local network environments with Graph2Vec (light blue), and binding sites with Bionoi-AE (orange)