SSF-DDI: a deep learning method utilizing drug sequence and substructure features for drug–drug interaction prediction

Table 7 Comparative evaluation in inductive setting (%)

Setting	Method	ACC	AUC	F1	Prec	Rec	AP
Random Split	CNN-DDI	70.64	82.95	61.61	89.1	47.11	83.79
	MR-GNN	75.99	84.85	72.3	85.52	62.68	84.89
	SSI-DDI	75.13	83.26	72.36	81.52	65.15	83.48
	GAT-DDI	77.94	86.58	75.28	85.63	67.16	85.81
	GMPNN-CS	79.95	89.34	77.22	89.33	68.02	89.25
	DGNN-DDI	77.07	87.35	73.03	88.08	62.07	86.97
	SSF-DDI(ours)	81.93	92.98	78.88	94.89	67.5	93.38
Structure-based Split	CNN-DDI	64.12	72.87	50.52	81.91	36.24	73.65
	MR-GNN	67.33	76.52	59.71	78.41	48.59	75.25
	SSI-DDI	68.52	77.41	62.06	78.63	51.43	77.14
	GAT-DDI	71.55	80.71	65.91	82.23	55.02	80.44
	GMPNN-CS	71.57	81.9	63.83	87.68	50.21	82.9
	DGNN-DDI	70.31	85.11	59.41	93.86	43.45	86.71
	SSF-DDI(ours)	77.22	85.93	71.96	93.55	58.46	88.23

ISSN: 1471-2105