Derek Nexus predicts human skin sensitisation accurately. What is the rationale behind its predictive performance?.pdf
The validation of new in vitro tests for skin sensitisation often involves comparing results against data taken from animal studies, typically the LLNA or GPMT assays. However, when considering the suitability of new tests for risk assessment it is important to also consider performance against human data since it is the potential for toxicity in humans that is ultimately of interest.
With this in mind, we decided to investigate how accurately Derek Nexus – an SAR-based expert-system for the prediction of toxicity designed and developed at Lhasa Limited – predicted a dataset of substances with well identified hazards for the endpoint of human skin sensitisation.
Derek Nexus showed satisfactory predictive performance against various human datasets collected from publically available sources (66-76% balanced accuracy) with high sensitivity (82-88%) but relatively low specificity (58-71%).
One explanation for this may be the fact that Derek Nexus alerts for skin sensitisation have been derived primarily using data from animal assays, such as the LLNA and the GPMT, which are often conservative tests and have a tendency to generate false positive predictions when measured against human data. Therefore it is possible that the alerts in Derek Nexus have inherited the conservative nature of these animal studies. Potential methods for improving predictions against human data are discussed.
Presented by Sebastien Guesne at 2014 EuroTox, Edinburgh, UK; 7th - 10th September 2014.