QSAR models for P-glycoprotein transport based on a highly consistent data set.

Area of Science:

• Pharmacology
• Computational Chemistry
• Biochemistry

Background:

• P-glycoprotein (Pgp) significantly influences drug elimination and disposition.
• Existing Pgp transport data is heterogeneous, hindering accurate predictive modeling.
• Standardized datasets are crucial for developing reliable Pgp interaction models.

Purpose of the Study:

• To establish a curated dataset of P-glycoprotein (Pgp) efflux ratios (ERs) in MDCK-MDR1 cells.
• To develop and validate predictive models for Pgp transport using diverse computational approaches.
• To identify key molecular descriptors and classification algorithms for accurate Pgp substrate prediction.

Main Methods:

• Compiled a dataset of 478 Efflux Ratios (ERs) in MDCK-MDR1 cell lines.
• Selected a subset of 187 compounds tested under consistent conditions in Borst-derived MDCK-MDR1 cells.
• Generated 23 predictive models using various descriptors, classification algorithms, and variable selection techniques.

Main Results:

• The most accurate predictive models achieved external validation accuracy of approximately 0.86.
• Four highly accurate models were based on VolSurf+ (VS+) descriptors.
• Two top-performing models employed Naïve Bayes (NB) classifiers with 4 novel descriptor selection techniques.

Conclusions:

• A robust dataset for Pgp transport modeling has been established.
• VolSurf+ descriptors and Naïve Bayes classifiers show high predictive power for Pgp substrates.
• The novel descriptor selection technique enhances the accuracy of Pgp interaction models.