Skip to main content
Fig. 5 | Journal of Cheminformatics

Fig. 5

From: Benchmark of four popular virtual screening programs: construction of the active/decoy dataset remains a major determinant of measured performance

Fig. 5

Correlations between the BEDROC scores (α = 80.5) and the properties of the targets and the small molecules. Upper panels: correlations with the protein properties, in orange, i.e., with the hydrophobicity of the cavity, its exposure and its number of atoms at 4 Å from the surface, in addition to the crystal structure resolution. Middle panels: correlations with the small molecule properties, in purple for the actives and green for the decoys, i.e., the molecular weight (MW), the octanol/water partition coefficient (AlogP), the electrotopological state (E-state), the polar surface area (PSA), the number of hydrogen bond acceptors (HBA), the number of hydrogen bond donors (HBD), the ring count (RC), the number of rotatable bonds (RB) and the embranchment count (EC). Lower panels: correlations of the BEDROC scores with the 2D fingerprints similarities between the actives (Sim2Dact-act) and between the actives and the crystal structure ligand (Sim2Dcryst-act), in blue. In all panels, the darkness of the colors depends on the p-value of the correlation, dark when p-value ≤10−4, medium light when 10−4<p-value ≤10−2 and light when p-value >10−2

Back to article page