Table 3.

In vitro methods used in combination for classifying and predicting skin sensitization potential of novel chemical compounds.

Combination methods	Description	Accuracy	Reference
(a) Peptide reactivity (b) Cell-based ARE† assay (c) TIMES-SS‡ computer modelling (d) Calculated octanol-water partition coefficient	• Scores of 0–4 for each individual test • A binary system is applied for in silico test results	88% (based on LLNA data) (116 test substances)	Natsch et al. (2009)
(a) DPRA (b) LuSens (similar principle with KeratinoSensTM assay) or KerotinoSensTM assays (c) h-CLAT or MUSST	• A sensitizer if DPRA and LuSens yield negative results and MUSST is positive • If contradictory results between DPRA and LuSens, or h-CLAT, then weight of evidence approach is used	94% (based on human data) 83% (based on LLNA data) (54 test substances)	Bauch et al. (2012)
Bayesian network Integrated Testing Strategy (a) TIMES§ (b) DPRA (c) ARE luciferase activity (d) MUSST	• Adaptive testing strategy where the choice and sequence of tests performed are based on available information • Reduces uncertainty of the sensitizing capacity of a test substance before proceeding to the experiment.	–	Jaworska et al. (2011)

^†ARE, antioxidant response element.

^‡TIMES-SS, tissue metabolism simulator for skin sensitization.

^§ TIMES, tissue metabolism simulator.