At this stage of the challenge, the goal is to identify primary hits. In the absence of prior ligand information for the target, we are aiming at exploiting the state-ofthe-art protein subpocket-fragment interactions for de novo design, via a 4-step workflow. In the first step, the sc-PDB database [1] (30,000 protein-ligand complexes in the 2022 archive) ligands are fragmented in their co-crystallized conformation with an interaction-aware fragmentation method [2]. Then, the protein environment around those fragments (shape and pharmacophoric features referred to as subpockets) [3] are compared to the target binding site, aligned, and scored with a local subpocket comparison method (ProCare) which had the advantage to be visually interpretable [4-5]. Fragments originating from the most similar subpockets are positioned into the target binding site according to their subpocket alignment. In the second step, fragments are assembled into larger molecules by designing linkers with a 3D deep generative method (DeLinker) [6], resulting in a target-focused library. In the third step, molecules are filtered according to desirable physicochemical properties (drug-likeness, synthetic accessibility) and searched in available/on-demand commercial databases. In the final step, available compounds are assessed by docking and rescoring using both empirical scoring functions [7-8] and topological validated methods (IChem) [9]. Identified hits from the first-round design will be grown using the above-described procedure, focusing on unexplored subpockets in the target protein cavity. The unique workflow proposed here has been validated in previous published and unpublished studies, enabling (i) off-target prediction of approved drugs supported by experimental validation (see reference 5), and (ii) design of a target-focused library among which molecules were similar to know inhibitors and new hits were identified and validated experimentally.

References:

[1] Desaphy, J.; Bret, G.; Rognan, D.; Kellenberger, E. Nucleic Acids Res. 2015, 43 (D1), D399–D404.

[2] Desaphy, J.; Rognan, D.J. Chem. Inf. Model. 2014, 54 (7), 1908–1918.

[3] Desaphy, J.; Azdimousa, K.; Kellenberger, E.; Rognan, D. J. Chem. Inf. Model. 2012, 52 (8), 2287–2299.

[4] Eguida, M.; Rognan, D. J. Med. Chem. 2020, 63 (13), 7127–7142.

[5] Eguida, M.; Rognan, D. J. Cheminform. 2021, 13 (1), 1–13.

[6] Imrie, F.; Bradley, A. R.; van der Schaar, M.; Deane, C. M. J. Chem. Inf. Model. 2020, 60 (4), 1983–1995.

[7] Korb, O.; Stützle, T.; Exner, T. E. J. Chem. Inf. Model. 2009, 49 (1), 84–96.

[8] Reulecke, I.; Lange, G.; Albrecht, J.; Klein, R.; Rarey. hemMedChem 2008, 3 (6), 885–897

[9] Da Silva, F.; Desaphy, J.; Rognan, D. ChemMedChem 2018, 13 (6), 507–510.

OpenEye Scientific Software (Santa Fe, NM 87508, U.S.A. https://www.eyesopen.com), Corina (Molecular Networks GmbH, 90411 Nürnberg, Germany. https://mn-am.com), HYDE (BioSolveIT GmbH, 53757 Sankt Augustin, Germany. https://www.biosolveit.de)

IChem (free academic license), ProCare, DeLinker, Protoss (free academic license), PLANTS (free academic license), RDKit

At this stage of the challenge, the goal is to predict active molecules among the provided set. In the absence of known selective ligands for the target protein whose PDB is available, we are aiming at performing a structure-based virtual screening via docking and rescoring. On the one hand, the target PDB will be prepared and protonated, optimizing the intramolecular hydrogen bonds. This task can be accurately performed with the Protoss software [1], which we have been using in previous published studies. On the other hand, the provided molecules will be filtered, with a particular attention to duplicates, and prepared by generating proper protonation states at physiological pH, considering potential tautomerization states. These tasks are achievable using the OpenEye Scientific Software suite [2] alongside with our customized inhouse filtering rules. Following conformers generation, the 3D molecules will be docked into the target central pocket with PLANTS [3], scored by the empirical scoring function ChemPLP [3] and rescored with HYDE [4], a scoring function based on hydrogen bond and dehydration terms. Docking poses can be quickly filtered or categorized according to their interactions with the target protein using the IChem toolkit [5]. The strategy described above was previously applied in our lab for hit identification projects, including for ultra high-throughput virtual screening of the Enamine REAL space. We think that our prior experience with this approach, combined to the available structural knowledge on the target and related proteins will be an asset to face this challenge. References: [1] Bietz, S.; Urbaczek, S.; Schulz, B.; Rarey, M. Protoss: A Holistic Approach to Predict Tautomers and Protonation States in Protein-Ligand Complexes. J. Cheminform. 2014, 6 (1), 12. https://doi.org/10.1186/1758-2946-6-12. [2] OpenEye Scientific Software, Santa Fe, NM 87508, U.S.A. https://www.eyesopen.com/ [3] Korb, O.; Stützle, T.; Exner, T. E. Empirical Scoring Functions for Advanced Protein−Ligand Docking with PLANTS. J. Chem. Inf. Model. 2009, 49 (1), 84–96. https://doi.org/10.1021/ci800298z. [4] Reulecke, I.; Lange, G.; Albrecht, J.; Klein, R.; Rarey, M. Towards an Integrated Description of Hydrogen Bonding and Dehydration: Decreasing False Positives in Virtual Screening with the HYDE Scoring Function. ChemMedChem 2008, 3 (6), 885–897. https://doi.org/10.1002/cmdc.200700319. [5] Da Silva, F.; Desaphy, J.; Rognan, D. IChem: A Versatile Toolkit for Detecting, Comparing, and Predicting Protein–Ligand Interactions. ChemMedChem 2018, 13 (6), 507–510. https://doi.org/10.1002/cmdc.201700505.

OpenEye Scientific Software (Santa Fe, NM 87508, U.S.A. https://www.eyesopen.com), Corina (Molecular Networks GmbH, 90411 Nürnberg, Germany. https://mn-am.com), HYDE (BioSolveIT GmbH, 53757 Sankt Augustin, Germany. https://www.biosolveit.de)

PLANTS (free academic license), IChem (free academic license)