Year: 2025
Type: article
Abstract: Background: Non-alcoholic steatohepatitis (NASH) is a chronic metabolic disease characterized by lipid accumulation, inflammation, and progressive liver fibrosis. The lack of approved treatments highlights the need to identify new agonists of the farnesoid X receptor (FXR), a key regulator of lipid metabolism and bile acids. Methods: An integrated AI-assisted drug discovery pipeline was developed, combining a QSAR model based on graph neural networks (GraphConvModel), high-throughput virtual screening, in silico ADMET evaluation, and molecular docking. The model was trained on 3,450 ligands from ChEMBL and PubChem, then applied to screen approximately one million additional molecules from PubChem. Results: The model achieved high performance (R² = 0.89), reproducing the experimental activity hierarchy. Successive filters (Lipinski, bioavailability, DILI, hERG) led to nine candidate agonists, with molecules M9 (CID: 7190831) and M1 (CID: 8888557) showing the best docking scores (−49.37 and −49.11 kcal/mol). Their stable interactions with residues PHE363, ILE359, TYR358, and TRP451 confirm their agonist potential. Conclusion: The AI-ADMET-Docking pipeline is a rapid and reliable predictive tool for discovering FXR agonists. These results validate the value of AI in researching innovative treatments for NASH and pave the way for experimental validation of molecule M9 (CID: 7190831). Keywords: FXR, NASH/MASH, molecular modeling, Artificial Intelligence, Graph Neural Networks (GNN)
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4416247355