The largest organ of the body is the skin, which serves as a protective layer against the presiding environment; however, skin can be breached by chemicals, which may have either a beneficial or deleterious impact on the body. Transdermal drug delivery (TDD) has been used in a beneficial manner to provide a therapeutic avenue for systemic treatment of a variety of maladies.

The basic premise of skin permeation is the penetration of a chemical through the stratum corneum (SC) dermal layer and absorption into systemic circulation. In general, the chemical has two potential routes of entry: (a) through hair follicles and sweat ducts, and (b) across the SC. Hair follicles and sweat ducts occupy only a small fraction of the total skin surface area and are believed to be insignificant factors in TDD. To permeate through the SC, the chemical must first partition into the skin and then diffuse through the hydrophilic and/or lipophilic environment of the SC to the dermis. The ability of a chemical to permeate through the skin is quantified by the value of the permeation coefficient, Kp, which is related to Fick's law of diffusion.

The ability to predict accurately the dermal permeation of a chemical is important for the development of new therapeutic formulations in transdermal drug delivery and assessment for potential risk of environmental chemicals. Most early modeling efforts utilized traditional and linear quantitative structure-property relationship (QSPR) models; however, since most thermo-physical properties have non-linear relationships with chemical structure, traditional linear algorithms result often in inferior QSPR models. New models accounting for all the structural features of importance to skin permeation, which avoid the pitfall of over-reliance on linear models, are required.

Our models for skin permeation, as expressed by the permeation coefficient, improve on other literature models in several aspects, including (a) the use of a larger dataset consisting of diverse chemical structures, (b) selection of a dataset consisting of both literature and statistically determined descriptors (c) use of multiple QSPR software to provide descriptors assuring both model superiority and stability, (d) use of nonlinear transformations to obtain the most suitable set of descriptors, and (e) use of robust non-linear neural networks with multiple randomizations and initializations to ensure network stability. Employing these state-of-the-art nonlinear algorithms, we have modeled accurately skin permeation with an absolute average deviation, root mean square error, and correlation coefficient of 8.2, 0.38, and 0.93, respectively. These results support our transdermal drug delivery research in the identification of candidate chemical penetration enhancers.