Chemistry Reference
In-Depth Information
scoring with diverse/large sets of complexes. With correlation to experiment (R
2
)
ranging 0.12-0.58, 19 different methods were submitted although no advantage
for any type of scoring function was observed.
Badly and well scored complexes had ligands of the same size although properties
(torsional and hydrogen bonding) varied significantly. It was suggested that
improvements in scoring should come from minimization of complex structures,
explicit account for water molecules, protein flexibility, more thermodynamic
accurate methods for calculating free energies of bonding, target-specific scoring,
curaton of dataset entries, paying attention to noncovalent interactions and
chemical features.
The publicly released 2012 Community Structure-activity resource (CSAR) data
set contains 508 compounds with known affinity values, 757 compounds with
provided SMILES strings, 57 compounds with an available protein-ligand crystal
structure, 185 compounds designated to be inactive. The compounds affinities,
which span six protein targets, are given in different units of measurement from
different assays. The data set provides the separated ligand and protein in their
native bound conformations, the complex compound and a set of unbound ligand
conformations [819].
The data sets were recently used to compare selected knowledge-based and force
field based scoring functions. The knowledge-based component seemed to
improve binding mode predictions. Scoring experiments, with full set of 2012
CSAR benchmark, indicates that without the knowledge of native protein and
ligand conformations it is challenging to predict affinities and binding modes.
The van der Waals scoring functions perform much better on native protein and
ligand conformations and can be used as a reference for scoring comparison. For
binding affinity predictions, the use of pregenerated ligand conformations seems
to lower the success rates much more than the use of non-native protein
conformations. There is a smaller difference for binding affinity predictions. This
may be due to the usage of side chain flexibility mainly for protein
conformational changes. Future trends involve usage of side chain rotamer
sampling and on-the-fly ligand conformational sampling.
