Improving protein-ligand binding site prediction accuracy by classification of inner pocket points using local features

Table 1 Datasets statistics

Dataset	Proteins	Ligands	#L	#P _FP	#P _CC	Cov _FP [%]	Cov _CC [%]	LS	PS _FP	PS _CC
CHEN11	251	476	1.90	12.41	1.75	71.0	52.3	26.9	38.9	51.0
ASTEX	85	143	1.68	21.58	2.25	81.1	65.7	23.2	41.9	56.9
DT198	198	192	0.97	18.57	2.19	80.2	65.6	20.8	41.2	53.7
MP210	210	288	1.37	14.50	1.99	78.8	68.2	22.8	40.0	50.9
B48	48	54	1.13	12.06	1.96	92.6	81.5	21.9	37.8	44.2
U48	48	54	1.13	11.40	1.79	88.9	77.8	21.9	38.0	46.8

Abbreviations: FP Fpocket, CC ConCavity.
#L: average number of ligands for one protein.
#P: average number of predicted pockets for one protein.
Cov: total coverage – success rate considering all predicted pockets (measured by D_CA criterion with 4 Å threshold).
LS: average number of heavy atoms in a relevant ligands (ligand size).
PS: average number of protein surface atoms that belong to a predicted pocket (pocket size).

ISSN: 1758-2946