Table 2. Prediction Performance of Proposed Pose Classification on CASF-2016^a.

		No. Poses	Accuracy (%)	Precision	Recall	F1 score
RF_i	incorrect	1633	69.9	0.70	0.96	0.81
	correct	796		0.68	0.16	0.26
3D-CNN	incorrect	1633	83.4	0.92	0.82	0.87
	correct	796		0.70	0.86	0.77
3D-CNN_i	incorrect	1633	83.7	0.84	0.93	0.88
	correct	796		0.82	0.65	0.72
3D-CNN_a	incorrect	1633	87.4%	0.87	0.96	0.91
	correct	796		0.89	0.71	0.79
3D-CNN_ia	incorrect	1633	88.4	0.93	0.89	0.91
	correct	796		0.80	0.87	0.83
PCN	incorrect	1633	81.4	0.86	0.86	0.86
	correct	796		0.71	0.72	0.72
PCN_a	incorrect	1633	82.1	0.89	0.84	0.86
	correct	796		0.70	0.79	0.74
PCN_ia	incorrect	1633	87.2	0.90	0.92	0.91
	correct	796		0.82	0.78	0.80

^aFrom top to bottom, Random Forest with protein–ligand interaction features (RF_i), 3D-CNN, 3D-CNN with protein–ligand interaction features (3D-CNN_i), 3D-CNN with affine transformation (3D-CNN_a), 3D-CNN with both interaction features and affine transformation (3D-CNN_ia), PCN, PCN with affine transformation (PCN_a), and PCN with both interaction features and affine transformation (PCN_ia).