Fig. 3: Interpretable analysis of DMFF-DTA model.

a A statistical table showing the average and standard deviation of attention weights output by the model across the Binding Site, Binding Range, and Outside of Binding Range regions for Davis and KIBA datasets. Below is the t-test result comparing attention distributions between different regions. “ns” indicates no significant difference, while asterisks denote significant differences (p < 0.05) between regions. b Visual cases of drug-target complexes from PDB entries 4G5J, 4XEY, and 6VNK. Each row shows one case, with a 3D visualization of the complex on the left indicating the binding pose. Red residues are those assigned high attention weights by the model. The middle column illustrates interactions between the target and drug molecule. Attention weights output by the model for the drug molecule are shown on the right, with deeper colors representing higher attention values.