Read‐Across Prediction of the Acute Toxicity of Organic Compounds toward the Water Flea Daphnia magna

According to the European REACH Directive, the acute daphnid toxicity needs to be assessed for industrial chemicals with market volumes ≥1 t/a. Employing a data set of 1365 organic compounds with experimental 48-h LC50 data for Daphnia magna, a read-across approach has been developed that makes use of the atom-centered fragment (ACF) method as quantitative measure for structural similarity. Both quantitative log LC50 predictions and a discrimination between narcosis-level and excess toxicity can be obtained, augmented by similarity-triggered information that characterizes a compound as inside or outside the quantitative or qualitative model domain. Reading across proceeds as interpolation of the toxicity enhancement (Te ) over predicted narcosis-level toxicity, taking experimental log Te values from similarity-selected reference compounds as input. The resultant decision tree model yields r(2) =0.85 and rms=0.66 for the subset of 757 compounds (56 %) identified as inside the quantitative model domain, and can handle further 318 compounds (23 %) with the categorical submodel, with 290 compounds (21 %) being outside its domain. The new in silico approach appears useful as ITS (Integrated Testing Strategy) tool for the daphnid toxicity assessment. The discussion includes a comparison of Kow - and LSER-predicted narcosis-level toxicity in the read-across context.