Machine learning-driven drug repurposing for GPR17: activity prediction via graph neural networks and multistage...

Area of Science:

• Computational chemistry
• Neuroscience
• Drug discovery

Background:

• G protein-coupled receptor 17 (GPR17) is involved in oligodendrocyte differentiation, neuroinflammation, and remyelination, making it a key target for multiple sclerosis (MS) therapy.
• Developing novel GPR17-targeting ligands is crucial for advancing myelin repair strategies.

Purpose of the Study:

• To identify novel GPR17-targeting ligands using an integrated computational workflow.
• To explore drug repurposing opportunities by analyzing learned chemical spaces.
• To provide computationally prioritized candidates for early-stage GPR17 ligand discovery.

Main Methods:

• Training a graph neural network (GNN) model on a dataset of 323 ligands with reported pIC50 values.
• Employing multistage structure-based validation, including pharmacophore-guided virtual screening and molecular docking.
• Utilizing molecular dynamics simulations and MM-PBSA/MM-GBSA calculations for binding energy assessment and ADMET profiling.

Main Results:

• The GNN model demonstrated robust predictive performance (R² up to 0.868).
• Analysis revealed enrichment of antihistamine and leukotriene scaffolds, suggesting drug repurposing potential.
• Several candidate molecules, including barmastine and sulukast, showed stable binding modes with high hydrogen-bond occupancy.

Conclusions:

• The hybrid computational workflow offers a practical strategy for early-stage GPR17 ligand discovery.
• The study proposes computationally prioritized candidates for further experimental validation in CNS-oriented drug development.
• This approach facilitates the identification of potential therapeutics for multiple sclerosis and other neurological disorders.