Linear discriminant analysis for skin sensitisation potential of diverse organic chemicals

摘要

Skin sensitisation is one of the emerging toxicological endpoints posing a significant concern for human health. Stimulation index (SI) and murine local lymph node assay (LLNA) are used to estimate quantitatively the skin sensitisation potential to classify chemicals as skin sensitising or non-skin sensitising. To obviate these time-consuming and expensive approaches, development of in silico predictive models has gained considerable attention over the last few decades. In this background, we have developed a linear discriminant analysis (LDA) model for LLNA based on the skin sensitisation potential of 147 chemicals with wide diversity of molecular structures. The developed LDA model is rigorously validated using various classification metrics, which show that the model is able to discriminate the skin-sensitising and non-skin-sensitising compounds. The developed model and contribution plot suggested that rotatable bond or molecular flexibility gives negative contribution, whereas dragon branching index gives positive contribution towards both skin-sensitising and non-skin-sensitising compounds. The descriptors such as number of sulphonate fragments (thio-/dithio-), number of triple bonds, number of nitrogen atoms and quadric index have equal contributions on the skin-sensitising and non-skin-sensitising property. Finally, the model was applied to screen DrugBank database compounds to identify the compounds that are likely to have skin sensitisation.