Structure-based machine learning model for discovering pregnane X receptor (PXR) agonists and biological activity...

Area of Science:

• Pharmacology and Computational Chemistry
• Nuclear Receptor Signaling
• Drug Discovery

Background:

• Pregnane X receptor (PXR) is crucial for bile acid homeostasis and inflammation, making it a therapeutic target for cholestatic liver diseases and inflammatory bowel disease.
• PXR regulates key metabolic enzymes (CYP3A4, UGT1A1) and transporters, and modulates inflammatory pathways (NF-κB).

Purpose of the Study:

• To develop a structure-based machine learning strategy for identifying novel PXR agonists from natural product databases.
• To validate the predictive power of the machine learning model and identify potent natural product-derived PXR agonists.

Main Methods:

• A novel structure-based machine learning approach integrating ligand-based and structure-based features using Light Gradient Boosting Machine.
• Pharmacophore modeling for initial screening, followed by machine learning prediction of PXR agonistic activity.
• In vitro validation using HepG2 cell culture and dual-luciferase reporter assays to determine EC50 values.

Main Results:

• The developed machine learning model achieved high accuracy (R²=0.874 internal, R²=0.845 external validation), outperforming other regression models.
• Schisantherin A, rhynchophylline, and irigenin were identified as potent PXR agonists with EC50 values of 1.58 μM, 2.57 μM, and 20.67 μM, respectively.
• The identified compounds demonstrate significant PXR agonistic activity.

Conclusions:

• The novel machine learning strategy effectively identifies potent PXR agonists from natural products.
• Schisantherin A, rhynchophylline, and irigenin are promising candidates for developing targeted therapies for PXR-related diseases.
• This work supports the design and discovery of new PXR modulators for therapeutic applications.