论文部分内容阅读
OBJECTIVE The aim of this study was to refine the prediction model to only give a binary Sensitizer (S)/Non-Sensitizer (NS) call without impacting the predictive capacities of the MUSST.METHODS In the present Myeloid U937 Skin Sensitization Test (MUSST) test method, co-stimulatory molecule CD86 as a marker of cell activation is measured.In parallel, cell viability is assessed using propidium iodide exclusion.Both parameters are measured by flow cytometry.208 substances with LLNA data were analyzed.MUSST results on this set gave 92 S, 56 NS predictions and 60 Inconclusive (INC).Statistical methods based on a stacking scoring method were applied on raw data of these 60 INC in order to find additional rules to classify these substances into S or NS.RESULTS A combination of 6 rules was identified, that allows to classify the INC into the 2 classes "sensitizers and non-sensitizers".Application of the rules to the 48 ECVAM dataset submitted in 2009 show an overall accuracy of 92% with a sensitivity of 96% and a specificity of 86%.The predictive performances of the MUSST on the 208 substances with the optimized prediction model composed of the original prediction model and the additional 6 rules-based model show 84% concordance, 87% sensitivity and 79% specificity against LLNA data.No false negative are observed among the extreme and strong sensitizers.CONCLUSION The prediction model of the MUSST classifies the substances as S or NS.However, it would be inappropriate to consider the MUSST assay as a stand-alone method for skin hazard characterization (CLP classification).The MUSST test method is modeling the dendritic cell activation.Such dendritic activation information is foreseen to play an important role in a future integrated approach for both hazard characterization and potency.