Development of a novel phase separating nail lacquer for the treatment on onychomycosis

Human skin permeability has been shown to be inherently non-linear when mathematically related to the physicochemical parameters of penetrants. These studies have also shown that non-linear methods, such as Gaussian processes (GPs), have outperformed other methods, such as quantitative structure–permeability relationships (QSPRs), in terms of predictivity and statistical accuracy. The aim of this study was to apply and validate GP methods to datasets for membranes other than human skin. Two QSPR methods were employed to compare with the GP models. As measures of performance, the correlation coefficient, negative log estimated predictive density and mean squared error were employed. GP models, with different covariance functions, outperformed QSPR models for human, pig and rodent datasets. For the Silastic membrane, GPs performed better in one instance, and gave similar results in other experiments. The GP predictions for some of the Silastic dataset were often poorly correlated, suggesting that the physicochemical parameters employed in this study might not be appropriate for developing models that represent this membrane.