Internal and external validation of the long-term QSARs for neutral organics to fish from ECOSAR (TM)

This study concentrates on the external validation of an existing Quantitative Structure-Activity Relationship (QSAR) model widely used for long-term aquatic toxicity to fish. In the context of the REACH legislation, QSARs are used as an alternative for experimental data to achieve a complete environmental assessment without the need for animal testing. The predictivity of the model was evaluated in order to increase the reliability of the model. We assessed whether the model met all of the OECD principles. The model was adapted to become more robust, and predictions were made with an external validation set collected from several databases. For the internal validation of the QSAR, the r(2), Q(LOO)(2) and Q(LMO)(2) were used as validation criteria, and for the external validation r(2), Q(ext)(2), h and the validation ratio were used. A few substances were classified as outliers and therefore the applicability domain of the QSAR had to be adjusted. The QSAR passed all validation criteria and met all the OECD principles for QSAR validation, and the long-term toxicity QSAR for fish can be applied with high certainty of a correct prediction within the limits of the inherent uncertainty of the model in cases where the substance falls within the applicability domain.