Classification of P-glycoprotein-interacting compounds using machine learning methods.

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Biochemistry

Background:

• P-glycoprotein (Pgp) is a key transporter involved in multidrug resistance and drug pharmacokinetics.
• Inhibiting Pgp is a crucial strategy for overcoming multidrug-resistant cancers and enhancing treatment efficacy.
• The complex nature of Pgp and experimental variability present challenges in identifying Pgp-interacting compounds.

Purpose of the Study:

• To develop and validate machine learning models for classifying Pgp-interacting compounds.
• To identify key physicochemical properties influencing Pgp inhibitor and substrate activity.
• To provide tools for the rational design and screening of novel Pgp inhibitors.

Main Methods:

• Classification of 2,477 Pgp-interacting compounds using machine learning algorithms: decision tree induction, artificial neural network, and support vector machine.
• Analysis based on physicochemical properties of the compounds.
• Model performance evaluation using internal cross-validation and external validation metrics (MCC values).

Main Results:

• Machine learning models demonstrated good predictive performance for classifying Pgp inhibitors, non-inhibitors, substrates, and non-substrates.
• Achieved high Matthews Correlation Coefficient (MCC) values, ranging from 0.739–1 for internal validation and 0.665–1 for external validation.
• Identified simple and interpretable models highlighting crucial properties influencing Pgp compound activity.

Conclusions:

• The developed machine learning models offer reliable classification of Pgp-interacting compounds.
• These models can aid in the efficient screening and rational design of Pgp inhibitors.
• The findings are significant for improving therapeutic outcomes in multidrug-resistant cancers.