Computer Regression Models for P-Glycoprotein Transport of Drugs

Abstract

Regression models of the cellular substrate specificity of 177 drugs for P-glycoprotein were built using linear regression, random forest, and support vector methods. QSAR modeling used a full-trial search of all possible combinations of the seven most significant molecular descriptors with clear physicochemical interpretations. The statistics of the obtained models were satisfactory according to an internal cross-validation and external validation tests using 44 new compounds. H-bond descriptors were components of almost all most significant QSAR models. This confirmed that H-bonds played an important role in penetration of the compounds through the blood–brain barrier. The developed statistical models could be used to assess P-glycoprotein transport of investigational new drugs.