AI-Driven Discovery Pipeline Identifies Carvedilol as a Novel Partial Agonist of THRβ

Thyroid hormone receptor beta (THRβ) is a key therapeutic target for metabolic dysfunction-associated steatotic liver disease and dyslipidemia, yet the discovery of structurally diverse agonists remains challenging. Here, we developed an AI-driven hierarchical drug discovery framework integrating graph neural network (GNN)-based virtual screening, structure-based modeling, and multilevel biological validation to accelerate THRβ agonist identification. The optimized GNN model outperformed conventional machine-learning approaches and enabled efficient prioritization of active compounds from an in-house library. Subsequent molecular docking and MM/GBSA rescoring identified Carvedilol as a high-confidence candidate. TR-FRET assays confirmed that Carvedilol acts as a partial THRβ agonist with preferential SRC2-2 coactivator recruitment, comparable to Resmetirom. Cellular assays further demonstrated that Carvedilol significantly reduced lipid accumulation and triglyceride levels and upregulated key metabolic regulators, including DIO1, ME1, and LDLR. Molecular dynamics and dcTMD analyses elucidated that Carvedilol is stabilized within the ligand-binding pocket through dominant hydrophobic interactions and specific hydrogen bonds with Thr273 and His435. Overall, this study repurposes Carvedilol as a structurally distinct THRβ agonist and validates an effective AI-guided strategy for nuclear receptor-targeted drug discovery.