GOLD: Library Predictions

This page gives a summary of two preliminary studies into the use of GOLD to dock combinatorial libraries.

• Acids and amines
• Sulphonyl chlorides and amines
• Comments

Acids and Amines

Method

A set of commercially available carboxylic acids was selected and converted to MOL2 format by Dr R C Glen (Tripos). Similarly, a set of amines was chosen. These may be regarded as two "monomer pools". Calculations were performed to determine whether GOLD might be used to select a subset of the acids and a subset of the amines, such that the resulting combinatorial library contains amides that are likely to be active.

The method chosen was to dock each of the acids and each of the amines into the protein of interest, and then determine which pairs of acids and amines were docked in such a way that an amide link could be formed between them in a reasonable geometry. If this were the case, it was considered that the resulting amide was likely to be an active compound.

In more detail, the procedure was as follows. A program was written to convert each acid to the corresponding methylcarbamoyl derivative (RCO₂H -> RCONHMe) and each amine to the corresponding acetylamino derivative (R'NH₂ -> R'NHCOMe). The resulting molecules were docked separately by GOLD into the active site of the enzyme lipase (PDB code 1CRL). Three dockings were performed on each monomer. However, if the first two solutions were within 1.0A RMSD, then the third solution was not generated. In view of the simplicity of most of the monomer structures, and in order to speed things up, a reduced GA parameterisation was used. Specifically, each GA run comprised 50000 operations only and the initial energy barriers were increased (this parameterisation has been verified on the standard GOLD test set of 100 PDB complexes). Finally, the methyl group terminating the amide bond of each docked molecule was deleted and replaced with a dummy atom.

A C program was written to compare each docked acid moiety with all the docked amine moieties. When it was possible to superimpose the linking amide-group atoms of a docked acid and a docked amine to within an RMSD of 1.5A, it was considered that the two components might form a product amide which could bind to the protein (the superposition included the dummy atom of each monomer and the corresponding carbon atom in the other monomer). A second screen was then applied to remove high energy compounds. Specifically, if the van der Waals energy of the product was greater than 500 kcal/mol, then the compound was rejected. After elimination of duplicates, this resulted in 129 "docked" products.

Results

All CPU timings are on an R4400 SGI Indigo II.

In order to test the success of this approach, the amides identified through the above superposition procedure were docked separately by GOLD, as complete molecules. The reduced GA parameterisation was used again, though 5 dockings were generated. The best of these was compared with the prediction obtained using the superposition method. When the discrepancy between the two exceeded 3.0A RMSD, the amide was docked again using the normal, full GA parameterisation to see if a better fit could be obtained.

This docking verification stage took 209 CPU hours.

The table below summarises the RMSD between the docked orientations obtained by joining together the separately docked monomers, and the orientations predicted by normal GOLD docking of the complete product amide:.

Sulphonyl Chlorides and Amines.

A similar set of calculations was performed on the same set of amines and a set of sulphonyl chlorides. As before, the "most active" product sulphonamides were predicted by: (i) docking the separate monomers; (ii) identifying pairs of amines and sulphonyl chlorides that were positioned in such a way that atoms of the sulphonamide linkage could be superimposed with an RMSD of 1.5A or less; (iii) eliminating "products" with high-energy conformations (> 500 kcal/mol).

Following the superposition stage, there were 369 pairs of sulphonyl chlorides and amines overlaid. After application of the energy screen, there were 348 compounds. Many of these were duplicates, obtained by using different solutions of the same monomer. These duplicates were removed by selecting the superposition whose monomer solutions had the largest combined GOLD score. Following this procedure, there were 190 unique superpositions (i.e. 190 sulphonamides predicted to be "active").

Results

As before, the sulphonamides predicted to bind by the above procedure were then docked separately by GOLD. This verification stage was done in two passes of independent dockings, as described above for the acids and amines.

The table below summarises the RMSD's between the docked positions obtained from the monomer superposition method and the position obtained by GOLD docking of the complete sulphonamide.

Comments

We can conclude that a combinatorial library could, in principle, be designed by using GOLD to dock the monomers rather than the complete library of products. This is computationally tractable and therefore a methodology worth further investigation. Obviously, the ultimate proof of the method requires synthesis and testing of the designed library - something not yet done.

The method works better on the amides rather than the sulphonamides; presumably, this is related to the torsional flexibility of the linkage.

Back to main GOLD page