We will use a novel method named SpaceDock that we are currently benchmarking, which consists in two steps :(i) docking 150K commercial building blocks (e.g. amines, acids, sulfonylchlorides, etc...) in the binding site of interest, (ii) assemble pairs of chemically-compatible building blocks on-the-fly according to a set of 40 common organic chemistry reactions (e.g. amide bond formation, Suzuki cross-coupling, reductive amination, etc..) and in-house developed geometric rules (distances/angles) enabling the selection of appropriate pose pairs.
Preliminary trials with 5K known ligands indicate that docking the corresponding commercial building blocks (leading in a single step to the target ligand with one of our 40 organic chemistry reactions) is indeed feasible and leads to accurate ligand poses (rmsd < 2 Ang. in 80% of test cases, unpublished data). We have assembled a set of 125K building blocks commercially available and preselected by Enamine (Y. Moroz, Enamine, personal communication) for generating the REAL space, thereby maximizing the chance that the final product is really synthesizeable. All building blocks will be docked into the 8GCT structure according to a previously determined optimal building block docking procedure (GOLD docking, ChemPLP scoring), 20 poses will be saved for every building block, therefore enabling 400 recombination/ligand. After recombination, the pose will minimized with OpenEye SZYBKI in the binding site, using the MMFF94 forcefield and finally rescored according to 3 different scoring functions (ChemPLP, HYDEscore, protein-ligand interaction fingerprint-IFP simialrity to that of the 8GCY reference). Hits will be ranked by full IFP similarity to 8GCY first, then by polar IFP similarity to 8GCY (accounting for h-bonds and ionic bonds only), and finally by increasing HYDEscore. The best 10K hits will clustered by maximal common substructures to yield a representative/cluster for purchase and synthesis.