Selection of Appropriate Training Set of Chemicals for Modeling Dermal Permeability Using Uniform Coverage Design

In silico approaches to model dermal permeability require a training set of chemicals. Selection of an appropriate set of chemicals is the key factor in the precision of a model. The objective is to develop a training set of chemicals representing a wider chemical space relevant to biological activity such as dermal permeability and compare it with the currently used training set of chemicals by our laboratory, which was based on a subjective selection process. Wider chemical space refers to structurally diverse chemicals with a wide range of all the descriptor values. A parent dataset of 4098 chemicals with 5 solvatochromic descriptors obtained from ADME boxes database was used for this study. The approach for diverse chemical selection was performed using ‘uniform coverage design’ (UCD) run by SpaceFill program and compared with a cluster analysis using SAS. Five sets of 25 chemicals were obtained from the design and evaluated based on gas chromatographic assay amenability and representation of structurally diverse group. A final set of 25 chemicals to be used for modeling dermal permeability was selected, based on aforementioned criteria. Graphical plot of the principal components demonstrated that currently used training set of chemicals have narrow representation of parent dataset whereas the selected training set have appropriate representation in terms of chemical space. QSAR models were built from both training set of chemicals. The model based on the 25 selected chemicals (R2=0.83) performed equally if not slightly better then the model based on currently used training set of 32 chemicals (R2=0.82).