The GIVA cPLA2 enzyme, found in the cytosol, is implicated in various inflammatory diseases. Synthetic inhibitors targeting GIVA cPLA2 have gained significant interest as potential anti-inflammatory drugs.
Design Strategies:
Utilize the ChEMBL database, which contains over 340 records with relevant data on biological activity, for ligand-based virtual screening.
Employ a structure-based design approach using the crystal structure of human cytosolic phospholipase A2.
Schematic Representation of Proposed Workflow:
Key Steps:
Eliminate compounds with unfavorable properties through the following methods:
Apply substructure filters to remove PAINS.
Use filters based on non MedChem-friendly SMARTS.
Remove compounds with excessively high lipophilicity (cLogP) predicted by Molsoft software.
Conduct virtual screening guided by a consensus of ligand-based methods and docking.
Select structurally diverse compounds using the Min-Max algorithm.
Summary of Ligand-based Strategy:
Filter the stock collection based on predicted lipophilicity values (cLogP) from ICM models.
Utilize classification machine learning models as an initial virtual screening filter:
The ChEMBL database, containing 340 records of known cPLA2 inhibitors with relevant biological activity data, is used for training data. Decoys in the training set are generated using the DUD-E methodology, and decoys in the test set are obtained from ChEMBL.
Two independent ML models are used: Random forest built on count-based Morgan fingerprints and XGBoost model built on ECFP4 fingerprints.
The scores from the above models are combined.
Select compounds with enhanced chemical space coverage through diversity picking.
The ligand-based selection from the stock collection results in approximately 100,000 diverse compounds ready for structure-based studies.
Structural Information:
The PDB database contains only one structure (uniprot ID P47712, PDB code: 1CJY) with the desired domains (CAP/Lid).
The PDB structure of the Lid region (code: 1CJY) contains unresolved loops and side chains. These regions were modeled using the FREAD approach. Additional molecular dynamics studies were conducted to improve sampling of these regions.
Key Steps:
Perform a 10 ns molecular dynamics simulation using the OpenMM engine in explicit solvent to enhance sampling of the Lid/CAP region.
Detect the binding site using the ICM algorithm, which appears stable during the simulation.
Cluster the molecular dynamics trajectory based on RMSD values of the CAP region for subsequent docking studies. The three most populated clusters are selected.
Carry out molecular docking and scoring using Flare (Cresset).
In addition, conduct electrostatic complementary calculations using Flare (Cresset) as a structure-based approach.
Summary of Structure-based Strategy:
Perform a 10 ns molecular dynamics simulation to sample binding site conformations.
Employ consensus docking to the three most populated clusters generated by the molecular dynamics simulation.
Utilize electrostatic complementary calculations.
Select compounds based on docking and electrostatic complementary calculations, resulting in approximately 17,000 compounds with the best scores.
Further select diverse compounds through diversity picking, resulting in 5,600 compounds.
The distributions of basic physicochemical properties and representative structures are depicted in the following slides.
