TABLE 3.

Overall performance comparison of SPGNN and our BCGP method on lncRNA-miRNA-circRNA association prediction task.

Methods	lncRNA-miRNA				circRNA-miRNA
	F1	AUC	AP	NDCG	F1	AUC	AP	NDCG
SPGNN (Wang et al., 2023b)	0.430	0.894	0.419	0.828	0.403	0.840	0.510	0.911
GCNFormer (Yao et al., 2024)	0.305	0.677	0.226	0.686	0.350	0.730	0.453	0.815
BCGP (w/o NCN)	${0.432}^{†}$	${0.901}^{†}$	${0.432}^{†}$	0.808	${0.492}^{†}$	${0.919}^{†}$	${0.577}^{†}$	${0.929}^{†}$
BCGP (w/NCN)	0.439 $†$	0.903 $†$	0.435 $†$	0.812	0.572 $†$	0.948 $†$	0.712 $†$	0.957 $†$

The best results of four evaluation metrics (F1, AUC, AP, and NDCG) are highlighted in bold. In each dataset, significant improvements over the base model are marked with † (paired t-test, p < 0.05$).