Large-scale comparison of machine learning methods for profiling prediction of kinase inhibitors

Table 4 Performance comparison results of different graphs-based DL models on the test sets

Molecular feature	Method	AUC^g	F1 score^h	BAⁱ
Molecular graphs	GCN^a	0.729 ± 0.206	0.658 ± 0.271	0.604 ± 0.127
	GAT^b	0.675 ± 0.225	0.636 ± 0.272	0.582 ± 0.145
	MPNN^c	0.658 ± 0.202	0.621 ± 0.298	0.557 ± 0.128
	Attentive FP^d	0.674 ± 0.207	0.661 ± 0.295	0.581 ± 0.116
	Chemprop^e	0.717 ± 0.173	0.640 ± 0.291	0.573 ± 0.108
	FP-GNN^f	0.704 ± 0.223	0.627 ± 0.367	0.604 ± 0.142
	Mean	0.693 ± 0.028	0.641 ± 0.016	0.584 ± 0.018

ISSN: 1758-2946