Challenge #3

Hit Identification
Method type (check all that applies)
Deep learning
High-throughput docking
Description of your approach (min 200 and max 800 words)

Our proposal is to link a set of fragments co-crystallized with nsp3, using either a deep generative model or a knowledge-based linker database, to afford drug-like molecules spanning at least two subpockets of the target. In a first step, 186 co-crystallized fragments (Schuller et al., Sci Adv. 2021) occupying one of 5 possible subpockets (adenine site, oxyanion site, upper site, lower site, pyrophosphate loop site) will be placed in the atomic coordinate frame of a reference nsp3 structure (PDB ID 6W02). For each fragment, possible connectable atoms (with other fragments) will be selected manually and tagged. The second step will consist in linking fragment pairs according to their location. Only fragments occupying two neigbouring subpockets (adenine-oxyanion, adenine-upper, adenine-pyrophosphate, oxyanion-pyrophosphate, oxyanion-lower, pyrophosphate-lower) will be allowed to be connected to yield fully enumerated compounds. Two linking approaches will be pursued using either a deep generative model (Imrie et al. J Chem Inf Model, 2020) or a novel in-house organic chemistry-driven protocol enabling to link remote fragments directly the target 3D coordinate frame by a set of 5854 3D linkers derived from PDB fragmented-ligands (Eguida et al. J Med Chem, 2022). This new approach is followed by a quick minimization inside of the target-cavity (MMFF94 force field) to relax the structure and discard linkers leading to a modification of the seed fragment locations. All proposals will be docked to the reference nsp3 structure to ensure that at least one docking pose is able to locate the starting two fragments close to their X-ray coordinates (rmsd < 2 Å). After filtering for physicochemical, drug-likeness and synthetic accessibility properties (Eguida et al. J Med Chem, 2022), remaining hits will be searched for maximum common substructure similarity to REAL database compounds (Enamine Ltd, Kyiv, Ukraine). Up to 100 of the most similar compounds will be purchased.

What makes your approach stand out from the community? (<100 words)

The proposed approach relies on a unique workflow able to generate 3D molecules into a target-cavity by connecting prepositioned fragments of known binding mode, using a set of standard organic chemistry reactions (e.g. amide bond formation, Suzuki coupling, reductive amination). According the chemical environment of fragments atoms selected for pairing, the distance between connected atom pairs and proper organic chemistry rules are used to either directly connect two fragments by a single bond, or to link them by an appropriate linker derived from already fragmented ligand. The approach will enumerate drug-like and synthetically feasible compounds directly in the 3D target space.

 

Method Name
POEM (Pocket oriented elaboration of molecules)
Commercial software packages used

SYBYL x2.1.1, Certara USA Inc., Princeton, U.S.A.

Szybki, Filter: OpenEye Scientific Sofware, Santa Fe, U.S.A.

Corina: Molecular Networks GmbH, Nürnberg, Germany

Free software packages used

IChem, DeLinker, rdkit, POEM, PLANTS

Relevant publications of previous uses by your group of this software/method

Eguida, M., Schmitt, C., Hibert, M., Villa, P. and Rognan, D. (2022). Target-Focused Library Design by Pocket-Applied Computer Vision and Fragment Deep Generative Linking. J. Med. Chem., 65, 13771-13783

Da Silva, F., Desaphy, J. and Rognan, D. (2018) IChem: A Versatile Toolkit for Detecting, Comparing and Predicting Protein-Ligand Interactions, ChemMedChem, 13, 507-510

Desaphy, J. and Rognan, D. (2014) scPDBFrag: a database of protein-ligand interaction patterns for bioisosteric replacements. J.Chem. Inf. Model., 54, 1908-1918

Virtual screening of merged selections
Method type (check all that applies)
High-throughput docking
Description of your approach (min 200 and max 800 words)

The merged selection of potential hits will be docked to the ADP-ribose-bound nsp3 X-ray structure (PDB ID 6W02) using the PLANTS docking software (Korb et al. J Chem Inf Model, 2009). Caution will be taken to follow the rules recently described to enable reliable fragment docking (Schuller et al., Sci Adv. 2021), notably the explicit consideration of four strongly-bound water molecules. The set of 2D ligand structures provided by the CACHE organizers will be converted into 3D coordinates using the Corina software (Molecular Networks GmbH, Germany) and ionized at physiological pH using Filter (OpenEye Scientific Sofware, USA). Up to 4 enantiomers will be generated for compounds not bearing more than 2 stereocenters. Docking will be performed with the PLANTS software using standard settings (search speed 1, ChemPLP scoring function), keeping up to 10 poses differing by at least 1 Å rmsd. All docking poses will be rescored according to the protein-ligand interaction fingerprint (IFP) similarity (Marcou et al. J Chem Inf Model, 2007) to any of the 186 co-crystallized fragments used in the hit identification step. Docking poses will be further selected at the condition that IFP similarity to at least two fragment X-ray poses occupying two different subpockets is higher than 0.60.  Compounds verifying this IFP filter will be considered as active, others will be tagged as inactive.

What makes your approach stand out from the community? (<100 words)

The originality of the approach stems from the usage of in-house pioneered protein-ligand interaction fingerprints (Marcou et al., J Chem Inf Model, 2007) to select proper poses of potential hits. Hits will be considered as actives if their interaction pattern to the ADP-ribose cavity of nsp3 is similar enough to at least two of the previously co-crystallized fragments (Schuller et al. Sci Adv, 2021) at the condition that the latter are bound to any of two different subsites (adenine site, oxyanion site, upper site, lower site, pyrophosphate loop site). This selection protocol ensures that the hit occupies at least two neighboring subsites and thereby provides enough non covalent interactions to the target in order to increase their binding affinity with respect to the reference low affinity fragments.

 

Method Name
PLANTS/IChem
Commercial software packages used

Corina: Molecular Networks GmbH, Nürnberg, Germany.

Filter: OpenEye Scientific Sofware, Santa Fe, U.S.A.

Free software packages used

PLANTS: https://github.com/discoverdata/parallel-PLANTS

IChem: http://bioinfo-pharma.u-strasbg.fr/labwebsite/download.html

Relevant publications of previous uses by your group of this software/method

Marcou G. and Rognan, D. (2007) Optimizing Fragment and Scaffold Docking by Use of Molecular Interaction Fingerprints. J. Chem. Inf. Model, 47, 195-207.

Da Silva, F., Desaphy, J. and Rognan, D. (2018) IChem: A Versatile Toolkit for Detecting, Comparing and Predicting Protein-Ligand Interactions, ChemMedChem, 13, 507-510

Rivat, C., Sar, C., Mechali, I., Dioufoulet, L., Leyris, J.P., Sonrier, C., Philipson, Y., Lucas, O., Maillé, S., Haton, H., Venteo, S., Mezghrani,A., Joly; W., Mion, J., Schmitt, M., Pattyn, A., Marmigère, F., Sokoloff, P., Carroll, P., Rognan, D. and Valmier, J. (2018) Inhibition of neuronal FLT3 receptor tyrosine kinase alleviates peripheral neuropathic pain in mice. Nat. Commun., 9, 1042

Tran-Nguyen, V.-K., Bret, G. and Rognan, D. (2021) Accuracy of fast scoring functions to predict high-throughput screening data from docking poses: The simpler the better. J. Chem. Inf. Model., 67, 2788-2797