Table 1.

Electron density-based validation of protein–ligand models

Scores			Classification		Remedy
RSCC	% of structures	Predicted number of PDB	Twilight	VHELIBS
1.0–0.9	67	~46,900	Ligand fits density	‘Good’	Ligand model good to use
<0.9–0.8	21	~14,700	Ligand fits density partially	‘Dubious’	Ligand over-modeled or may benefit from further refinement
<0.8–0.7	7	~4,900	Significant parts of ligand not in density	‘Bad’	Use ligand model with caution!
<0.7–0.6	3	~2,100	Very poor fit of ligand to density
<0.6–0.5	1	~700	Improbably poor fit of ligand to density, density almost absent
<0.5	1	~700	Catastrophic fit of ligand to density, density completely absent

Real Space Correlation Coefficient (RSCC) values were calculated using Twilight [12, 15] and classifications determined using Twilight and VHELIBS [14]. The percentage of structures was determined using a curated set of 382,588 ligands from the PDB with electron density available in the EDS [20]. Pre-calculated data available for download from http://bit.ly/1shcwu4. The predicted number of PDB entries was calculated using the percentage of ligands in each category in the curated set and applying to the total PDB count (just over 70,000 protein–ligand models in the PDB as of Nov, 2014). This is likely a conservative estimate based on the assumption that each PDB files contains just one ligand