The new CCDC/Astex test set consists of 305 protein-ligand complexes. All protonation states have been assigned by manual inspection. It is a considerably extended version of the original GOLD validation test set.

The new set has been thoroughly checked, and a subset of ideal entries has been identified for use as a reliable validation suite. The Protein Data Bank (PDB) entries for both the full and clean sets can be found here.

Filters that have been applied to derive the clean subset are:

• Filter I: Assessment of the involvement of crystallographically-related protein units in ligand binding. GOLD docking results are shown below for PDB entry 1mtw, for which crystallographically-related residues need to be taken into account:

Figure 1. GOLD results for PDB entry 1mtw. Docking of the native ligand in the binding site leads to a wrong solution (left), yet inclusion of the crystallographically-related protein residues (middle) results in an acceptable solution (right).

• Filter II: Identification of bad clashes between protein side chains and the ligand. An example of a severe clash in PDB entry 1srf is shown below:

Figure 2. Examples of a crystal structure with very short contacts between the ligand and protein atoms: PDB entry 1srf, dC-C=2.2Å

• Filter III: Check for structural errors, and/or inconsistency of ligand placement with crystal structure electron density. The mismatch of electron density and ligand in PDB entry 1b6n is shown below:

Figure 3. PDB entry 1b6n; the ligand in the PDB lacks a phenyl sulfonamide group. Analysis of the original paper revealed that it had not been included in the PDB entry.

In addition, the set has been pruned to assure diversity in terms of protein-ligand structures.

The results have been investigated using the new test set of structures for both SuperStar and GOLD.