Table 3.

Performance of predicting drug‒drug interactions via the GIN of the group graph and other baseline models

Tasks	BIOSNAP			DrugBankDDI
Metric	AUC_ROC↑	PRC_AUC↑	F1↑	AUC_ROC↑	AUC_PRC↑	F1↑
SimNN	0.8530 ±0.0010	0.8480 ±0.0010	0.7140 ±0.0010	0.7860 ±0.0030	0.7530 ±0.0030	0.7090 ±0.0040
DeepDDI	0.8860 ±0.0010	0.8710 ±0.0070	0.8170 ±0.0070	0.8440 ±0.0030	0.8280 ±0.0020	0.7720 ±0.0060
CASTER	0.9100 ±0.0050	0.8870 ±0.0012	0.8430 ±0.0050	0.8610 ±0.0050	0.8290 ±0.0030	0.7960 ±0.0070
PretrainGNN	0.9948 ±0.0002	0.9939 ±0.0001	0.9607 ±0.0022	0.9716 ±0.0003	0.9668 ±0.0004	0.9172 ±0.0007
GIN(FGS-pretraining)	0.9957 ±0.0005	0.9940 ±0.0008	0.9795 ±0.0010	0.9792 ±0.0002	0.9755 ±0.0006	0.9343 ±0.0010
GIN(group graph)	0.9962 ±0.0002	0.9950 ±0.0003	0.9793 ±0.0015	0.9865 ±0.0004	0.9842 ±0.0007	0.9498 ±0.0011

The average values and 95% confidence intervals from three independent runs are reported