Graph neural network screen repurposes carvedilol as a partial THRβ agonist, confirmed in assays
- A hierarchical pipeline chained a graph neural network screen, docking with free-energy rescoring, and molecular dynamics to look for structurally new THRβ agonists relevant to fatty liver disease.
- The tuned network beat conventional machine-learning baselines at ranking actives from an in-house compound library.
- Carvedilol emerged as a candidate; biochemical coactivator-recruitment and cell-based lipid assays supported partial agonism similar to the clinical comparator resmetirom, with simulations pointing to key pocket contacts.
💡 Points Based on the abstract
Carvedilol is an old beta-blocker, so a THRβ hit is intriguing but needs scrutiny. I'd want to see selectivity against THRα and cardiac liability, plus whether the activity needs concentrations far above those reached in patients. The GNN's gain over baselines also depends on the split; scaffold-based splitting would show whether it generalizes beyond the library.
📄 Paper · Journal of Chemical Information and Modeling (ACS) ↗ AI-Driven Discovery Pipeline Identifies Carvedilol as a Novel Partial Agonist of THRβ