P-glycoprotein substrate models using support vector machines based on a comprehensive data set.

Area of Science:

• Biochemistry
• Pharmacology
• Computational Chemistry

Background:

• P-glycoprotein (P-gp) is a crucial ABC transporter involved in drug transport and cellular processes.
• Predicting P-gp substrates is vital for drug development, but reliable computational models are limited.

Purpose of the Study:

• To develop and validate computational models for predicting P-glycoprotein substrates.
• To identify molecular descriptors and substructures that differentiate P-gp substrates from nonsubstrates.

Main Methods:

• Utilized a dataset of 332 distinct chemical structures, including 131 substrates and 81 nonsubstrates.
• Employed ADRIANA.Code, MOE, and ECFP_4 fingerprint descriptors for molecular representation.
• Developed support vector machine models with rigorous cross-validation (5-, 10-fold, LOO).

Main Results:

• The best model achieved a Matthews Correlation Coefficient of 0.73 and a prediction accuracy of 0.88 on the test set.
• ECFP_4 fingerprint analysis highlighted substructures like nitrile and sulfoxide groups as significant for P-gp interaction.
• Structural isomerism in sugars demonstrated a notable impact on P-gp substrate likelihood.

Conclusions:

• The developed computational models offer a reliable approach for predicting P-gp substrates.
• Identification of key functional groups and structural features can guide the design of compounds with specific P-gp interaction profiles.
• This work contributes to improving drug efficacy and reducing adverse effects by better understanding P-gp transport mechanisms.