Original GOLD test set
A simple test of the effectiveness of a docking program is to take a protein-ligand complex from the
Brookhaven Protein Data Bank and extract the ligand.
The docking program can then be used to predict the binding mode of the ligand and a comparison made with the
crystallographically observed position. This methodology has been used to validate GOLD.
Tests were done in two phases: first, on a test set of 100 complexes; later, on an additional 34
complexes as a check against over-training. Results of these two test phases are summarised below.
Results of experiments on an initial dataset of 100 PDB complexes.
These experiments are described in:
G. Jones, P. Willett, R. C. Glen, A. R. Leach & R. Taylor, J. Mol. Biol.,
267, 727-748, 1997
[10.1006/jmbi.1996.0897]
and were first presented at
Molecular Interactions, the 15th Annual Conference and AGM of the Molecular Graphics
and Modelling Society, University of York, UK, 16-19 April 1996.
The method used for each test calculation was as follows. Parts of the
protein remote from the binding site were deleted, in order to speed up
the calculation. Enough of the protein was retained to ensure that
all residues were present that might reasonably interact with the ligand.
The ligand was extracted from the protein binding site. Hydrogens were
placed on both the protein and the ligand in order to ensure that
ionisation and tautomeric states were defined unambiguously. Obviously,
this sometimes involved making hypotheses about the protonation states
of residues such as His, Glu and Asp. The ligand was minimised into a
low-energy conformation and the atom types of both the protein and
ligand checked for accuracy.
In almost all test runs, all water
molecules were deleted from the protein structure.
This is not strictly defensible since
water molecules often mediate protein-ligand binding. However, if
more careful judgements were made on which waters to remove,
the effect would be to improve the accuracy of the GOLD predictions.
Hence, the deletion
of all waters is a conservative strategy which will make GOLD look less
reliable than it really is, rather than more reliable.
20 GA runs were performed on each test complex, using the default GA
parameter settings.
Results of experiments on a subsequent dataset of 34 PDB complexes.
The GOLD algorithm was improved in various ways following the above set of validation tests.
Some time later, therefore, a second set of tests was performed on 34 additional complexes in
order to ensure that GOLD had not been over-trained on the original set.
A file, original_set.tar.gz, containing all test set
files can be downloaded from our website.
Pictures of dockings generated by GOLD.
| 
