using simplified scoring functions may accelerate calculation at the
expense of lower accuracy.
The accuracy of docking is often quantified by Root Mean
Square Deviation (RMSD) between a docking predicted structure
and an experimentally determined one. However, most experimen-
tally determined structures are not available in practice. Alterna-
tively, molecular 3D visualization system, such as Pymol, VMD
[12], and Chimera [13], offers intuitive inspections of docking
results. To confirm the prediction, interface shape complementar-
ity, hydrophobicity, hydrogen bonding, and other relevant interac-
tions are checked by visualization. Besides, using independent
scoring functions is a shortcut for docking evaluation [9, 14]. In
contrast to the scoring functions in docking programs, these inde-
pendent functions are more accurate but time-consuming. They are
comprised of more energy terms or employ sophisticated models to
fit the experimentally determined binding affinities. A number of
independent scoring functions have been proposed in recent years
[15–20]. Benchmark assessments showed their power in binding
affinity prediction and virtual screening [21].
Cyscore is an independent protein–ligand scoring function
developed in our group [15]. It mainly focus on improving the
prediction of hydrophobic free energy, which is dominant in most
protein–ligand binding [22–25]. Conventionally, hydrophobic free
energy is treated as surface tension, which is proportional to inter-
facial surface area for simplicity and efficiency. However, this model
ignores the role of molecular shape and hinders the accurate
prediction. Inspired by the thermodynamic research of Tolman
[26] and the pioneer work of Nicholls et al. [27, 28], Cyscore
takes curvature as the descriptor of shape and quantifies the hydro-
phobic free energy by a curvature dependent surface area model.
Figure1 illustrates the evaluation of docking by Cyscore on a
106 protein–ligand set, which was generated by a well-known
docking software, DOCK6. The differences between predicted
and experimentally determined binding structures are quantified
by RMSD. The Cyscore of docking results versus RMSD illustrates
the relationship between docking accuracy and Cyscore. The figure
shows that the RMSD is lower than 2 A ̊ when Cyscore is smaller
than3.5. It indicates that Cyscore is an indicator for evaluating
protein–ligand docking.2 Materials
The input of Cyscore includes a protein file in PDB format (see-
Notes 1and 2 ) and a ligand file in SYBYL MOL2 format. The
ligand file is obtained from the output of a docking program such as
DOCK6. Hydrogen atoms are required for Cyscore calculation.
Cyscore package includes Cyscore, CurvatureSurface, RotaBond,
and AddH.234 Yang Cao et al.