Fig. 2

Cumulative performance boost of either regression or classification type problems was attained using the SEF descriptors. a Comparison of network performance with cumulative addition of different Shannon entropies in the descriptor set. Ki values of binding molecules to the human coagulation factor 11 were analyzed using the metric R² of fit (%). b The addition of the Shannon (SMILES) entropy to the descriptor set consisting of MW and BEI of ligands (ligands BEI) improved the overall performance of the deep neural network. The scaling factors of metrics were listed in Additional file 1: Table S6. c The cumulative increase in ROC_AUC and accuracy of the toxicity classification of Ames mutagenicity dataset by cumulative addition of different Shannon entropy-based descriptors. The used descriptor sets were 1. Shannon (SMILES), 2. fractional Shannon (SMILES), 3. fractional Shannon (InChiKey), 4. Shannon (SMILES) + Shannon (SMARTS) + Shannon (InChiKey) + fractional Shannon (InChiKey) + bond freq, and 5. Other descriptors + Shannon (SMILES) + fractional Shannon (SMILES). The other descriptors were listed in Additional file 1: Table S8. All prediction metrics were averaged over at least 5 independent runs