Development of a read-across method for evaluating repeated-dose toxicity using molecular descriptors and in vitro test data

Abstract

Repeated-dose toxicity (RDT) test is important for the safety assessment of chemical substances. Meanwhile, the development of non-animal RDT evaluation methods is highly demanded. In this study, we aimed to improve read-across accuracy for RDT using in vitro test data in addition to chemical structure information. Male rat 28-42-day RDT test data of 326 substances were obtained from Hazard Evaluation Support System Integrated Platform (NITE, Japan). By grouping similar endpoints (EPs), 6 hepatotoxicity-related and 2 hematology-related group EPs (gEPs) were defined, and if a substance was positive for any EP in a gEP, the substance was considered positive for the gEP. We used 2649 descriptors from alvaDesc (Alvascience), and the results (positive/negative) of rat cytochrome P450 inhibition tests and HepG2 cell-based cytotoxicity tests and high-content analyses. Thirty-two substances were selected as test substances, and their neighbor (i.e., similar) substances were selected from the rest of 294 substances based on Euclidean distance between substances. Sensitivity (0.5 - 1.0), specificity (0.1 - 0.7), and balanced accuracy (0.6 - 0.8) differed significantly among gEPs. Next, we selected an in vitro test parameter most associated with each gEP by Fisher's exact test, and only the neighbors whose results matched those of test substances were used for read-across. This selection improved prediction accuracy for some gEPs (hepatomegaly, dyslipidemia, and anemia). These results suggest that using gEP-associated in vitro test data can improve read-across accuracy for some gEPs.