Refining the prediction of skin sensitisation potential: performance of the Derek Nexus knowledge base for a published dataset
This poster was presented by Martin Payne at the 13th Congress of the European Society of Contact Dermatitis.
There are several structure-activity relationship-based in silico tools available for predicting skin sensitisation, including free and commercial software, e.g. Toxtree, the OECD QSAR Toolbox, VEGA, Times-SS, Case-Ultra, and Derek Nexus. Several publications recently have reported the performance of these systems.
Teubner et al. utilized results for 100 chemicals, derived largely (74/100) from on-line registration dossiers submitted to and compiled by the European Chemicals Agency (ECHA), and for the remaining chemicals from recent publications and in-house data. They reported a sensitivity of 93% and a specificity of 47%. More recently Kostal and Voutchkova-Kostal working with the same dataset, but with some revisions, reported a 90% sensitivity and 38% specificity for Derek Nexus. We considered the specificity in both publications low and examined the performance ourselves.
Teubner W, Mehling A, Schuster PX et al. (2013), Regul. Toxicol. Pharmacol. 67, 468-485.
Kostal J and Voutchkova-Kostal A (2015), Chem. Res. Toxicol., 29, 58-64.