Pesticides are substances widely used in agriculture to control pests, including weeds. One of the most informative and commonly used endpoint in regulatory settings is the acute oral toxicity in rat. This endpoint can be predicted using QSAR models on different computational platforms. These are typically global or similarity-based models. Classifications by modes of action (MOA) could generate local models with higher accuracy. In this study, we present a classification model by MOA of pesticides to estimate acute oral toxicity in rat. The IUPAC - Pesticides Property DataBase (PPDB) was used as a reference. IUPAC-PPDB contains 2288 pesticides including insecticides, fungicides and herbicides and are classified by MOA. Two databases were assigned to the most common MOA: the EPA T.E.S.T. database (v. 4.2.1), which contains 7413 pesticides, and pesticides used in Mexico (Cofepris Pesticides Database: CPD, which contains 150 compounds). Depending on the MOA, assignments were performed using Random Forest or KNN algorithms. An initial exploration of the databases consisted of the analysis of the chemical space coverage of the three databases (PPDB, T.E.S.T and CPD).  Predictive models, using these data, are in progress. As an example, we present a model of acute oral toxicity in rat, developed using 121 similar pesticides. Descriptors were calculated with Dragon 6. Predictive models (MLR), developed in QSARINS, consisted of multiple linear regressions, estimated with ordinary least squares, and genetic algorithms for variable selection. Leave-one-out cross validation, Y-scrambling and applicability domain was analyzed. The statistics of the best model are: Model development and internal validation: R2=0.816, Q2LOO=0.777. External validation: Q2F1=0.647 Retx=0.644, CCC=0.789.

Classification models of pesticides by mode of action

2019 Spring National Meeting

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