Exploring 3D-QSPR models of human skin permeability for a diverse dataset of chemical compounds

The control of permeation is vital not only for the topical application of lotions, creams, and ointments but also for the toxicological and risk assessment of materials from environmental and occupational hazards. To understand the effects of physicochemical properties of a variety of 211 compounds on skin permeability, we developed a three-dimensional quantitative structure-property relationship (3 D-QSPR) model. Alignment free GRid-INdependent Descriptors (GRINDs), which were derived from molecular interaction fields (MIFs) contributed to the regression models. Kennard–Stone algorithm was employed to split data set to a training set of 159 molecules and a test set of 52 molecules. Fractional factorial design (FFD), genetic algorithm (GA) and successive projection algorithm (SPA) were applied to select the most relevant 3 D molecular descriptors. The descriptors selected using various feature selection were correlated with skin permeability constants by partial least squares (PLS) and support vector machine (SVM). SPA-SVM model gave prominent statistical values with the correlation coefficient of R2= 0.96, Q2= 0.73 and R2pred=0.76. According to the analysis results, the hydrogen bonding donor and acceptor properties of the investigated compounds can influence the penetration into the human skin. Furthermore, it was found that permeability was enhanced by increasing the hydrophobicity and was diminished by increasing the molecular weight. In addition, the presence of hydrophobic groups in the target molecule, as well as their shape and position, can affect the skin permeability.