A decision tree-based integrated testing strategy for tailor-made carcinogenicity evaluation of test substances using genotoxicity test results and chemical spaces.

摘要: Genotoxicity evaluation has been widely used to estimate the carcinogenicity of test substances during safety evaluation. However, latest strategies using genotoxicity tests give more weight sensitivity; therefore, their accuracy very low. For precise evaluation, we attempted establish an integrated testing strategy for tailor-made materials, considering relationships among results (Ames, in vitro mammalian and vivo micronucleus), chemical properties (molecular weight, logKow 179 organic functional groups). By analyzing toxicological information 230 chemicals, including 184 carcinogens Carcinogenicity eXperience database, a decision tree was optimised statistically. A forest model generated machine-learning method-random forest-which comprises thousands trees. As result, balanced accuracies cross-validation model, space (71.5% 75.5%, respectively), were higher than example regulatory (54.1%). Moreover, statistical optimisation tree-based models revealed significant groups that would cause false prediction standard non-genotoxic (e.g., amide thioamide, saturated heterocyclic fragment aryl halide). In most important parameters all models, even when silico integrated. Although external validation is required, findings established herein will contribute determine new mechanistic hypotheses carcinogenicity.