In Silico Regression Modeling and Improved Interpretability To Predict the Transport Inhibitory Activity of Breast...

Area of Science:

• Pharmacology and Drug Discovery
• Computational Chemistry
• Medicinal Chemistry

Background:

• Breast cancer-resistance protein (BCRP) is a key efflux transporter involved in drug-drug interactions (DDIs).
• In silico methods, including binary classification models, are used to predict BCRP inhibition in early drug discovery.
• There is a need for regression models that provide quantitative predictions and enhanced interpretability for BCRP inhibitors.

Purpose of the Study:

• To develop and interpret regression models for predicting IC50 values of BCRP inhibitory activity.
• To identify key chemical substructures responsible for BCRP inhibition using interpretable machine learning.
• To screen approved drugs for potential BCRP inhibitory activity.

Main Methods:

• A high-quality dataset of 870 compounds with IC50 values was compiled from public sources.
• Initial regression models were built, achieving an R² of 0.736, but faced interpretability challenges.
• Novel machine learning models utilizing substructure-based fingerprints and tree-based algorithms were developed, incorporating SHapley Additive exPlanations (SHAP) for feature interpretation.

Main Results:

• The best regression model achieved an R² value of 0.736 on the test set.
• An interpretable model using substructure fingerprints achieved an R² of 0.703 on the test set.
• SHAP analysis identified key substructures, and seven approved drugs were predicted as potential BCRP inhibitors.

Conclusions:

• Interpretable regression models can accurately predict BCRP inhibitory activity.
• The identification of key substructures aids in understanding structure-activity relationships for BCRP inhibition.
• These predictive models can accelerate compound screening, reduce development costs, and identify potential drug candidates or liabilities.