Predicting toxicities of ionic liquids in multiple test species – an aid in designing green chemicals

Abstract

Ionic liquids (ILs) due to their unique characteristics have attained much importance for future applications, although they may pose environmental risks to aquatic ecosystems that have to be assessed. This paper presents a novel computational approach for estimating the toxicity of ILs in multiple test species of different trophic levels in accordance with the OECD guidelines. Here, ensemble learning based global structure–activity relationship (SAR) models were established for qualitative (two- and four-category) and quantitative toxicity predictions of ILs in Vibrio fischeri and successfully applied to the algae and daphnia species. Diversity and nonlinearity of the considered datasets were evaluated using the Tanimoto similarity index, Kruskal–Wallis and Brock–Dechert–Scheinkman statistics. The gradient boosted tree (GBT) and bagged decision tree (BDT) SAR models were constructed using simple descriptors and validated by stringent statistical tests. In V. fischeri, algae and daphnia data, the classification accuracies rendered by GBT and BDT models were 84.44–100% (two-category) and 92.23–98.74% (four-category), while the two models yielded correlations (R²) of 0.857–0.982 between the measured and predicted toxicity values with mean squared errors of 0.21–0.04. The SARs also identified structural elements of ILs responsible for their toxicities. The successful results obtained in three test species of different trophic levels reveal that the proposed approach can be useful as a screening tool to easily aid, from the early stages of the processes, the design of aquatic environmentally friendly ILs.