Table 1.
DDIMDL dataset.
Fig 1.
Overall framework of the proposed model.
The changes in the node colors indicate the process of node learning, and H(l) and Z(l) indicate the vector representations learned at the lth layers of the DNN model and GCN model, respectively.
Table 2.
Four different combinations of drug pairs.
Table 3.
Performance of our model against competitive approaches.
Table 4.
Ablation variants settings.
Fig 2.
Comparison of the ablation experiment results.
Fig 3.
Statistical analysis of the DDI class distribution.
Table 5.
Comparison of the link prediction results obtained over multiple classes.
Fig 4.
Distributions of the attention values over different classes.
(A) Attention distribution for Class_65. (B) Attention distribution for Class_30. (C) Attention distribution for Class_10.
Fig 5.
Experimental results in different tasks.
(A) Task A. (B) Task B.
Fig 6.
Results of ablation experiments.
(A) Effect of the number of GCN layers. (B) Effects of drug combination methods. (C) Effect of the fusion coefficient.
Table 6.
Case study prediction results.
Fig 7.
Top 100 pairs of DDI prediction results.