Cho_2016_Chemosphere_159_178

For designing environmentally friendly ionic liquids (ILs), their structural effects on the toxicity should be interpreted via modelling based on the quantitative-structure-activity-relationship (QSAR) concept. For the purpose, QSAR models for predicting IL toxicity in acetylcholinesterase activity were developed by using linear free-energy relationship (LFER) descriptors, whose chemical meanings are well defined. These are excess molar refraction (Ec or a), dipolarity/polarizability (Sc or a), H-bonding acidity (Ac or a), H-bonding basicity (Bc or a), McGowan volume (Vc or a), and ionic interactions of cation (J+) and anion (J-). Since the experimentally determined LFER descriptors are not available, we calculated them based on density functional theory, conductor-like screening model and the open-source software, obprop. The toxicity values of imidazolium- and pyridinium-based ILs could be predicted by a combination of four descriptors (Ac, Bc, Vc and Sa) with an R2 of 0.828, and (Ec, Ac, Ea and Sa) with an R2 of 0.879, respectively. In prediction study using the overall dataset containing various IL structures, the six calculated terms (Ec, Sc, Ac, J+, Ea, and Sa) were selected and correlated with the observed toxicity values in R2 of 0.748 for the training set, R2 of 0.711 for the test set and R2 of 0.655 for external validation set. And this study explains how the selected terms are contributing to the prediction models, and their chemical meanings were understood.