Table.

The best performance of the 4 models for the multilevel framework. The AUC, Recall, and Precision of the M-PvsP model were calculated with the weighted method.

Model	Feature	Algorithm	5-CV				Test
			ACC	AUC	Recall	Precision	ACC	AUC	Recall	Precision
B-PvsC	ECFP4	LR	0.933	0.974	0.974	0.946	0.955	0.981	0.990	0.957
B-MvsP	MACCS	LightGBM	0.858	0.916	0.908	0.865	0.843	0.902	0.869	0.874
M-PvsP	MACCS	LightGBM	0.783	0.904	0.783	0.757	0.809	0.932	0.809	0.794
B-McoL	2D	LightGBM	0.941	0.980	0.961	0.927	0.963	0.982	0.981	0.944