Submission Details
Molecule(s):
O=C(NC1=CC=CC[C@H]1C(=O)NC1=CC=CC1)Nc1cccnc1
O=C(NC1=CC=C(F)C1)Nc1ccc(CNc2ccccc2)cc1
O=C(Nc1ccccc1)c1cc(Cl)ccc1CNC1=CC=CC1
O=C(NC1=CC=C(F)C1)Nc1ccc(C(=O)Nc2cn[nH]c2)cc1
Design Rationale:
The molecules are designed using a Graph-Based Genetic Algorithm developed by Jan H Jensen (https://pubs.rsc.org/en/content/articlelanding/2019/SC/C8SC05372C) and modified by me to work with Glide docking. We use the PDB: 6LU7 structure as suggested on this site where the covalently bound ligand was removed and the CYS had a hydrogen atom added accordingly. We use the Zinc database to start from, but otherwise the algorithm just optimises for better binders without human intervention. The algorithm uses 50 generations to generate 400000 molecules of which 4 are submitted. We use functional group filters from the ChEMBL database to remove unlikely molecules when they are generated and use a synthetic accesibility heuristic to remove particularly hard-to-synthesize molecules. Out of the final ~ 8000 molecules we use ASKCOS to remove molecules once again that are particularly hard to synthesize and the final 44 molecules were re-docked with Glide using eXtra precision (XP) and the submitted molecules had a glide score between -8 and -7.6 kcal/mol. The suggested fragments listed below are just referring to the binding site. Compared to our last submission (https://covid.postera.ai/covid/submissions/751a2458-4d70-431d-ad9e-3151895919bc, https://covid.postera.ai/covid/submissions/167c18e3-3388-4802-bfa0-f8dc23d34258), higher precision was used for the genetic algorithm. This submission includes compounds that are predicted to be good binders but with lower synthetic availability than our previous submissions.
Other Notes:
Our ASKCOS and molecule.one predictions of synthetic availability states that these molecules are perhaps more difficult than our previous submissions.