Fuzzy modeling of skin permeability coefficients

The purpose of this work was to determine whether a new modeling methodology using fuzzy logic c an predict skin permeability coefficients that are given compound descriptors that have been proven to affect percutaneous penetration.

Three fuzzy inference models were developed using subtractive clustering to define natural structures within the data and assign subsequent rules. The numeric parameters describing the rules were refined through the use of an Adaptive Neural Fuzzy Inference System implemented in MatLab. Each model was evaluated using the entire data set. Then predicted outputs were compared to the published experimental data.

All databases produced fuzzy inference models that successfully predicted skin permeability coefficients, with correlation coefficients ranging from 0.83 to 0.97. The lowest correlation coefficient resulted from a model using log octanol/water partition coefficient and molecular weight as inputs with two input membership functions evaluated by two fuzzy rules. T he correlation coefficient of 0.97 occurred when log octanol/water partition coefficient and hydrogen bond donor activity were used as inputs with three input membership functions evaluated by three fuzzy rules.

Fuzzy rule-based models are a realistic and promising tool that can be used to successfully model and predict skin permeability coefficients as well as or better than previous algorithms with fewer inputs.