CGINet: graph convolutional network-based model for identifying chemical-gene interaction in an integrated multi-relational graph

Table 2 Performance comparison of our models with baseline approaches

Model	Component	AUROC	AUPRC	AP@20	TIME
DeepWalk	CG-graph	0.830	0.811	0.733	–
Node2Vec	CG-graph	0.819	0.800	0.735	–
SVD	CG-graph	0.833	0.823	0.772	–
Laplacian	CG-graph	0.839	0.841	0.765	–
GCN-CG	CG-graph	0.855	0.830	0.742	2.2
GCN-total	Total graph	0.823	0.768	0.571	8.5
CGINet-1	Two subgraphs	0.901	0.872	0.770	2.4
CGINet-2	Two subgraphs, \(\lambda = 0.4\), \(\mu \in \left[ {0,1} \right]\)	0.927	0.898	0.765	2.9
CGINet-3	Two subgraphs, \(\lambda = 0.5\), \(\mu = 1\)	0.914	0.893	0.804	2.8

The values of each metric are average performance in terms of different random seeds. The results are average performance values for all interaction types. TIME denotes the average training time of each epoch and it is measured in hours. The best result of each performance index is boldfaced

ISSN: 1471-2105