Prediction of Farnesoid X Receptor Disruptors with Machine Learning Methods

Area of Science:

• Computational chemistry
• Pharmacology
• Drug discovery

Background:

• The farnesoid X receptor (FXR) is a key regulator of metabolic pathways.
• Dysfunctional FXR activity due to xenobiotic binding can lead to adverse effects.
• Accurate identification of FXR-binding xenobiotics is crucial for drug safety.

Purpose of the Study:

• To develop and validate predictive models for identifying potential farnesoid X receptor (FXR) binders.
• To identify key chemical substructures associated with FXR binding.
• To assess the impact of applicability domain analysis on model performance.

Main Methods:

• Utilized five machine learning algorithms.
• Incorporated eight molecular fingerprints and 20 molecular descriptors.
• Employed information gain and substructure frequency analysis for feature identification.
• Performed external validation and applicability domain analysis.

Main Results:

• The best classification model combined molecular descriptors and fingerprints, achieving an AUC of 0.83 on the test set and 0.92 on an external validation set.
• The top model demonstrated over 85% prediction accuracy on a second external validation set.
• Identified privileged substructures like benzimidazole, indole, and stilbene moiety as important for FXR binding.
• Applicability domain analysis significantly improved prediction accuracy.

Conclusions:

• Developed robust machine learning models for predicting FXR binders.
• The models can aid in the rapid screening of chemicals for potential FXR interaction.
• Findings contribute to understanding FXR-xenobiotic interactions and mitigating drug-induced side effects.