The Astex Non-native Set is described fully in

Protein-ligand Docking against Non-native Protein Conformers.
M. L. Verdonk, P. N. Mortenson, R. J. Hall, M. J. Hartshorn, C. W. Murray
J. Chem. Inf. Model., 48, 2214-2225, 2008.
[DOI: 10.1021/ci8002254]

The performance of docking packages is most often validated by docking a ligand back into its native binding site and calculating the root mean square distance (RMSD) for the docked ligand pose compared to the native ligand pose. Few validations assess the docking performance against non-native protein conformations, the aim of the Astex non-native set.

The non-native set contains 1112 non-native structures taken from the PDB and was built on the Astex Diverse Set and were selected and prepared using the following methodology:

• the structure's resolution must be <2.5 Å
• for each entry in the diverse set, all structures of the same target were retrieved where the structure is either an apo structure or a complex containing a different ligand
• for each target, every non-native structure was superimposed onto the corresponding structure in the Astex Diverse Set
• ligands were extracted from the proteins and classified as solvent, cofactor or ligand; binding sites with no ligands were flagged as apo structures
• in cases where protein side chains were disordered in the non-native structure, only one of the conformers was selected based on occupancy or protein-ligand clashes
• all structures were then visually inspected

The results of the validation showed that while docking performance against native structures is 80% (top ranked solutions within 2 Å RMSD, it is only 61% for non-native docking. If all solutions are considered, the docking performance increases to 91% for native docking and 71% for non-native.