Prediction of water-phosphatidylcholine membrane partition coefficient of some drugs from their molecular structures

Abstract

In this work, the phosphatidylcholine membrane-water partition coefficients (MA) of some drugs were estimated from their theoretical derived molecular descriptors by applying quantitative structure-activity relationship (QSAR) methodology. The data set consisted of 46 drugs where their log MA were determined experimentally. Descriptors used in this work were calculated by DRAGON (version 1) package, on the basis of optimized molecular structures, and the most relevant descriptors were selected by stepwise multilinear regressions (MLRs). These descriptors were used to developing linear and nonlinear models by using MLR and artificial neural networks (ANNs), respectively. During this investigation, the best QSAR model was identified when using the ANN model that produced a reasonable level of correlation coefficients (R_train = 0.995, R_test = 0.948) and low standard error (SE_train = 0.099, SE_test = 0.326). The built model was fully assessed by various validation methods, including internal and external validation test, Y-randomization test, and cross-validation (Q² = 0.805). The results of this investigation revealed the applicability of QSAR approaches in the estimation of phosphatidylcholine membrane-water partition coefficients.

Keywords::

• Partition coefficients
• artificial neural networks
• multiple linear regressions
• quantitative structure-activity relationships