You are here:
An evaluation of selected (Q)SARs/expert systems for predicting skin sensitisation potential
Citation:
Fitzpatrick, J., D. Roberts, AND G. Patlewicz. An evaluation of selected (Q)SARs/expert systems for predicting skin sensitisation potential. SAR AND QSAR IN ENVIRONMENTAL RESEARCH. Taylor & Francis, Inc., Philadelphia, PA, 29(6):439-468, (2018). https://doi.org/10.1080/1062936X.2018.1455223
Impact/Purpose:
Allergic contact dermatitis (ACD) is estimated to constitute about 10-15% of all occupational diseases. Predictive testing to characterise substances for their skin sensitisation potential has historically been based on animal models such as the Local Lymph Node Assay (LLNA) and the Guinea Pig Maximisation Test (GPMT). In recent years, EU regulations, have provided a strong incentive to develop non-animal alternatives. Significant progress has been made in developing and evaluating non-animal test methods. There have been efforts to develop and evaluate the utility of in silico models for skin sensitisation including local and global (Q)SARs as well as expert systems. In this study, we selected three different types of expert systems: VEGA (statistical), Derek Nexus (knowledge based), TIMES-SS (hybrid) and evaluated their performance using 2 large datasets of substances that had been assessed for their skin sensitisation potential in animal models. We considered a model to be successful at predicting skin sensitisation potential if it had at least the same balanced accuracy as the LLNA and the GPMT had in predicting the outcomes of one another, which ranged from 79% to 86% depending on the dataset. We found that none of the expert systems evaluated was able to achieve such a high balanced accuracy in their global predictions, with balanced accuracies ranging from 56% to 65%. However, for substances within the domain of TIMES-SS, balanced accuracies were found to be 79% and 82%, for the two datasets, in line with the animal data. The expert systems evaluated could be extended in light of the additional data collected as part of this study. The incorrect predictions offer new insights for how the existing alerts within these expert systems could be refined. These datasets also offer exciting opportunities for the development of new models.
Description:
Predictive testing to characterise substances for their skin sensitisation potential has historically been based on animal models such as the Local Lymph Node Assay (LLNA) and the Guinea Pig Maximisation Test (GPMT). In recent years, EU regulations, have provided a strong incentive to develop non-animal alternatives. Here we selected three different types of expert systems: VEGA (statistical), Derek Nexus (knowledge based), TIMES-SS (hybrid) and evaluated their performance using two large sets of animal data, one of 1249 substances from eChemportal (354 sensitisers and 895 non-sensitisers) and a second of 515 substances from NICEATM (329 sensitisers and 186 non-sensitisers). We considered a model to be successful at predicting skin sensitisation potential if it had at least the same balanced accuracy as the LLNA and the GPMT had in predicting the outcomes of one another, which ranged from 79% to 86% depending on the dataset. We found that none of the expert systems evaluated was able to achieve such a high balanced accuracy in their global predictions, with balanced accuracies ranging from 56% to 65%. However, for substances within the domain of TIMES-SS, balanced accuracies were found to be 79% and 82%, for the two datasets, in line with the animal data.
URLs/Downloads:
DOI: An evaluation of selected (Q)SARs/expert systems for predicting skin sensitisation potential
https://doi.org/10.1080/1062936X.2018.1455223