The importance of developing non-experimental procedures for estimating skin irritation potential of chemicals has been increasing as a result of concerns regarding animal welfare and laboratory expenses. A number of expert systems and quantitative structure-activity relationship (QSAR) models have been proposed in literature for predicting skin irritation potential of compounds. However, the application of these models to predict skin irritation require a priori estimates of several physiochemical properties such as, octanol/water partition coefficient, melting point, lipid solubility, aqueous solubility, surface tension and vapor pressure. Thus, prediction of skin irritation potential using these models often requires other models capable of estimating the physiochemical properties. Further, the literature QSAR models are limited to a particular class of compounds and thus, lack in universal applicability.

In this work we have developed a skin irritation QSPR model using rabbit Draize test data consisting of over two hundred compounds from various chemical classes. The effectiveness of using a combination of traditional, functional group and statistical descriptors to improve the predictive capability has been studied. Our QSPR model is capable of predicting the skin irritation potential of chemical compounds with an R2 of 0.83. External validation of the QSPR model was performed using data from the human-patch test.