The prediction of diverse biological endpoints such as toxicity by computer-aided tools is well established nowadays. These methodologies offer several benefits, allowing to obtain the predictions in a fast and economic way. Moreover, the reduction in animal testing and chemical substances use, make these methods economical, ethical and environmentally friendly. Validated models allow for accurate predictions. As a result, regulatory agencies around the globe, accept in silico predictions for defined endpoints for impurities generated in the industrial production of agrochemicals and pharmaceuticals. In this work, we present the development and validation of QSAR models of the LD50 acute oral toxicity in rat. Local models were generated using subsets of similar compounds and were confronted to global models. For example, a local model of benzimidazoles was validated by different coefficients and techniques such as across validation by Fischer coefficient (Q2 F1), across validation by leave-one-out and leave-many-out (Q2 LOO, LMO), correlation coefficient (R2), Y-scrambling, among others. This work provides a framework of comparison of local vs global models, and the coverage of the chemical space is discussed.

Local vs global models for the prediction of acute oral toxicity in rat

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.