Soheila Rezaei, Hassan Behnejad and Jahan B Ghasemi
The control of permeation is essential for the topical application of lotions, creams, and ointments, and the toxicological and risk assessment of materials from environmental and occupational hazards. This study developed a three-dimensional quantitative structure-activity relationship (3DQSAR) model to predict permeation of a variety of 210 compounds through human skin. Molecular descriptors were computed using a GRid Independent Descriptors (GRIND) approach. After variable selection via the genetic algorithm method, the 118 selected descriptors were correlated with skin permeability constants by PLS regression and support vector machine (SVM). Partial least squares regression (PLSR) and support vector regression (SVR) are two popular Chemometrics models that are being subjected to a comparative study in the presented work. Kennard-Stone algorithm was employed to split data set to a training set of 150 molecules and a test set of 60 molecules. Genetic algorithm (GA), as an influential linear tool, was used to certain the best and interpretative subset of variables for the predictive model structure. The best results were obtained by PLS regression with the correlation coefficient of R2=0.77 and SVM regression with the correlation R2=0.79. This strategy led to a final 3DQSAR model that presented Q2=0.61 and R2 pred=0.73. The obtained results revealed that the hydrogen bonding donor and hydrogen bonding acceptor of investigated compounds dramatically influences their ability to penetrate through human skin. Furthermore, it was found that permeability was enhanced by increasing hydrophobicity and lowered with increasing molecular weight.
Partagez cet article