Ghanem Ouahid Ben, Mutalib M I Abdul, Lévêque Jean-Marc, El-Harbawi Mohanad
Faculty of Chemical Engineering, Universiti Teknologi Petronas, Bandar Seri Iskandar, 31750, Tronoh, Perak, Malaysia.
Faculty of Chemical Engineering, Universiti Teknologi Petronas, Bandar Seri Iskandar, 31750, Tronoh, Perak, Malaysia.
Chemosphere. 2017 Mar;170:242-250. doi: 10.1016/j.chemosphere.2016.12.003. Epub 2016 Dec 5.
Ionic liquids (ILs) are class of solvent whose properties can be modified and tuned to meet industrial requirements. However, a high number of potentially available cations and anions leads to an even increasing members of newly-synthesized ionic liquids, adding to the complexity of understanding on their impact on aquatic organisms. Quantitative structure activity∖property relationship (QSAR∖QSPR) technique has been proven to be a useful method for toxicity prediction. In this work,σ-profile descriptors were used to build linear and non-linear QSAR models to predict the ecotoxicities of a wide variety of ILs towards bioluminescent bacterium Vibrio fischeri. Linear model was constructed using five descriptors resulting in high accuracy prediction of 0.906. The model performance and stability were ascertained using k-fold cross validation method. The selected descriptors set from the linear model was then used in multilayer perceptron (MLP) technique to develop the non-linear model, the accuracy of the model was further enhanced achieving high correlation coefficient with the lowest value being 0.961 with the highest mean square error of 0.157.
离子液体（ILs）是一类溶剂，其性质可以被改变和调整以满足工业需求。然而，大量潜在可用的阳离子和阴离子导致新合成的离子液体成员不断增加，这增加了理解它们对水生生物影响的复杂性。定量结构活性∖性质关系（QSAR∖QSPR）技术已被证明是一种有用的毒性预测方法。在这项工作中，σ-轮廓描述符被用于构建线性和非线性QSAR模型，以预测各种离子液体对发光细菌费氏弧菌的生态毒性。使用五个描述符构建线性模型，预测准确率高达0.906。使用k折交叉验证方法确定模型的性能和稳定性。然后将从线性模型中选择的描述符集用于多层感知器（MLP）技术来开发非线性模型，该模型的准确率进一步提高，相关系数高达0.961，最低均方误差为0.157。
