Elham Baher*, Poneh Ebrahimi, and Naser Darzi Naftchali
Faculty of science, Department of Chemistry, Golestan University, Gorgan, Iran.
A B S T R A C T
The most key factor in transition of drugs across the biological membrane is their solubility in water. In this study, a novel theoretical model was proposed for the prediction of drug solubility in the aqueous media (log 1/S) by using employing the molecular structure descriptors. The data set consists of 58 different drugs. Molecular descriptors were calculated and selected by genetic algorithm (GA) and stepwise-multiple linear regression (SMLR) methods. The selected descriptors with GA were: momentum of inertia, total molecular surface area, difference in CPSA, PPSA-3 atomic charge weighted PPSA, FPSA-3 fractional PPSA, count of H-donors sites, HA dependent HDSA-2/SQRT (TMSA) and kier shape index. The selected descriptors with SMLR were: FPSA-3, Kier shape index, randic index, XY shadow and count of H-donors sites. Then prediction of log 1/S as a criterion of drug delivery was accomplished by support vector machine (SVM) by using GA and SMLR selected descriptors, separately. By comparison of results obtained, it was concluded that the GA-SVM model was superior over other models. The statistical result presented a good generality and ability of SVM model in drug delivery property. The simplicity, reliability and high speed calculation are some advantages of the present work.
Keywords: log 1/S, support vector machine, genetic algorithm, quantitative structure-activity relationship