Host: The Japanese Society of Toxicology
Name : The 50th Annual Meeting of the Japanese Society of Toxicology
Date : June 19, 2023 - June 21, 2023
(Introduction)
The large effort to switch the in vivo evaluation assay of skin sensitization of chemicals to the alternative non-animal assays has been paid because EU banned animal tests for cosmetics. However, an alternative method of evaluating intensity has not been developed, although it is important for risk assessment. We have developed a machine learning (ML) based-model to predict EC3 values in Local Lymph Node Assay (LLNA EC3), which is used as an indicator of the intensity of skin sensitization, but the model had not been established for the practical use, such as explanatory, interpretability and conducting external validation. In this research, we developed ML model to predict LLNA EC3 values for the practical use.
(Method)
We used safety data of 143 chemicals, which were involved in the Defined Approach Guideline at OECD (No.497). We randomly divided them into training and external validation data with a 4:1 ratio and built a regression model with XGBoost. The objective variable was LLNA EC3 value, and the explanatory variables were in vitro test data (DPRA, KeratinoSensTM, h-CLAT), relating to the adverse outcome pathway of skin sensitization, chemical properties, and molecular descriptors and alert information obtained from QSAR ToolBox.
(Result and Discussion)
Using data of 29 chemicals as external validation data, we calculated predicted LLNA EC3 values using our model and compared with values from experiments. The coefficient of determination R2 value was 0.68. The results of three in vitro tests ranked high in variable importance in our model, suggesting a significant contribution to the predicted results. These results suggested that our model was useful for a practical use to predict the intensity of skin sensitization because our model had an explanatory and interpretability, and its performance was evaluated by external validation.
