In this study, we present an extension of recently published GALAS modeling methodology to develop QSAR models of aquatic toxicity accounting for mechanistic knowledge in terms of mode of toxic action (MOA) data. The endpoints considered for analysis were median lethal concentration of test compound in water (LC50) to fathead minnows (Pimephales promelas) and 50% inhibitory growth concentration (IGC50) to protozoan Tetrahymena pyriformis. The experimental data used for building the models involved 579 LC50 values for fishes obtained from PubChem database, as well as protozoan IGC50 data for 1093 compounds received from CADASTER project. Modeling was performed in a two-step approach consistent with baseline and excess toxicity concepts. In the first step, a global linear regression model was derived, describing baseline toxicity levels as a function of the compound's lipophilicity, expressed by octanol/water logP. The second step involved analysis of systematic deviations introduced by the baseline model in the local chemical neighborhood of the compound, in order to account for excess toxicity of reactive chemicals. External validation of the obtained models demonstrated good predictive power with RMSE under 0.7 log units for estimating toxicity in both species. A further boost in model performance was observed, when an substructure-based expert system was used to classify chemicals by their MOA, so that only the compounds with the same MOA were considered in local analysis. Furthermore, the described models comply with OECD validation principles and are fully compatible with automatic generation QPRF reports.

Utilizing mode of action data to improve prediction of aquatic toxicity

246th National Meeting (2013)

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