Quantitative prediction of skin sensitisation potency based on structural alert spaces
This poster was presented by Dr Martyn Chilton at the 2016 Society of Toxicology Annual Meeting.
It was also presented by Dr Donna Macmillan at the 2017 British Toxicology Society Annual Congress.
The prediction of skin sensitisation potency is demonstrated by using a similarity approach within a defined chemical space. Quantitative predictions are made by considering the similarity between compounds which contain the same structural alert, thus restricting the chemical space to those compounds which are believed to cause skin sensitisation via the same mechanism.
Skin sensitisation potency can be measured using the estimated concentration of chemical required to induce a 3-fold increase in activity over the baseline in the murine local lymph node assay, a value known as the EC3. In this work, an in silico k-Nearest Neighbours model was developed which quantitatively predicts the potency (EC3 value) of a query chemical. The predicted EC3 was calculated using the weighted average of EC3 values of up to 10 most similar compounds within the same mechanistic space e.g. chemicals activating the same Derek skin sensitisation alert.
The methodology was evaluated using a validation set of 46 compounds. The performance of the model was judged based on correctly predicting EC3 values within a factor of three or if the predicted and experimental values fell within the same ECETOC potency categories (extreme = <0.1%, strong = =0.1% and <1%, moderate = =1% and <10%, weak = =10% and =100%). 52% of the predictions fell within the correct ECETOC potency category, and 48% of the predictions fell within 3-fold of the experimental EC3 value. The approach proved to be conservative as 35% of the remaining predictions for both performance metrics were predicted as more potent than observed experimentally.