A Machine Learning-Based Prediction Platform for P-Glycoprotein Modulators and Its Validation by Molecular Docking

Area of Science:

• Pharmacology
• Computational Chemistry
• Machine Learning

Background:

• P-glycoprotein (P-gp) overexpression drives multidrug resistance (MDR) in cancer by effluxing chemotherapy drugs.
• This efflux limits therapeutic efficacy, particularly in resistant and refractory tumors.
• Targeting P-gp with inhibitors is a key strategy to overcome MDR.

Purpose of the Study:

• To develop a machine learning model for predicting P-gp substrates and inhibitors.
• To identify novel compounds that can modulate P-gp activity.
• To provide a tool for rapid screening of large chemical libraries.

Main Methods:

• Utilized machine learning strategies, including random forest feature selection and leave-one-out cross-validation.
• Trained and validated the model using a curated list of P-gp modulators from the ChEMBL database.
• Employed molecular docking to validate predictions of substrate and inhibitor compounds.

Main Results:

• The developed machine learning model demonstrated high performance scores on an external validation set.
• Predicted P-gp substrates showed docking poses similar to doxorubicin.
• Predicted P-gp inhibitors exhibited docking poses comparable to the known inhibitor elacridar.

Conclusions:

• The machine learning approach is a valid and effective tool for identifying P-gp substrates and inhibitors.
• This method facilitates the rapid screening of large chemical libraries for potential MDR modulators.
• The findings support the development of new strategies to combat multidrug resistance in cancer therapy.