Biotoxicity Study Of Ionic Liquids Based On Machine Learning Algorithms

Ionic liquids(ILs)are room-temperature molten salts composed of anions and cations,and are widely used in many fields as a new alternative to organic solvents.With the massive discharge of ILs in the environment,the harm to the aquatic environment and human health has gradually become prominent.Therefore,there is an urgent need to assess the ecotoxicity of ILs.In this study,machine learning modeling approach combined with regression model was used to construct quantitative structure-activity relationship models(QSAR)for the toxicity of ILs to Daphnia magna,Zebrafish and Raphidocelis subcapitata.And interspecies quantitative structure-toxicity-toxicity(i-QSTTR)models were developed to evaluate the cross-species toxicity mechanism of ILs.Finally,in order to evaluate the potential effects of ILs on humans,we performed a systematic analysis of h CES2 inhibition by23 ILs using machine learning algorithms and molecular docking.The main work included:Five machine learning methods,including multiple linear regression(MLR),partial least squares regression(PLS),random forest regression(RF),support vector regression(SVR)and extreme gradient boosting(XGBoost)were used to develop models for predicting the toxicity of ILs to Daphnia magna,Zebrafish and Raphidocelis subcapitata.Rigorous validation criteria were used to verify the robustness and predictive performance of the models,and the results proved that the nonlinear SVR and XGBoost models outperformed the linear models.Subsequently,we analyzed the key molecular descriptors affecting the toxicity of ILs,and found that the structure of both anions and cations had an effect on the biological toxicity of ILs,and cations were the most critical factor affecting the toxicity of ILs.For example,the partition coefficient,molecular flexibility,and polarity of cations determined the toxicity of ILs to Daphnia magna.The toxic effect of ILs on Zebrafish mainly depended on the polarizability and the number of heavy atoms of cations.The molecular flexibility and van der Waals surface area of the cation influenced the toxicity of ILs to Raphidocelis subcapitata.The applicability domain(AD)of the models were characterized by the Euclidean distance-based method and Williams plot.Finally,the established optimal models were used to predict the biological toxicity of a series of ionic liquids with unknown toxicity,which effectively filled the data gap.It is of great significance to study the toxicity correlation and toxicity difference of ionic liquids to different species for the safe application of ILs.Interspecies quantitative structure-toxicity-toxicity(i-QSTTR)models were developed using toxicity data of ILs towards Daphnia magna,Zebrafish,and Raphidocelis subcapitata.The results showed that the toxicity of ILs to Daphnia magna and Zebrafish had a good correlation,indicating that ILs had the same or similar toxicity mechanism to these two organisms.At the same time,there was a good correlation between Zebrafish and Raphidocelis subcapitata,indicating that ILs had similar toxic effects on Zebrafish and Raphidocelis subcapitata.In contrast,the toxicity of ILs to Daphnia magna and Raphidocelis subcapitata was poorly correlated,suggesting that the toxicity mechanism of ILs to these two aquatic organisms were significantly different.In this study,the toxicity of ILs to one species was used to estimate the toxicity of other species,which provided theoretical support for further exploring the toxic mechanism of ILs.Finally,we evaluated the effects of ILs on humans.This study was based on h CES2 inhibition toxicity data for 23 ionic liquids.We modeled and trained the h CES2 inhibition toxicity of ILs using five machine learning algorithms.The modeling results demonstrated that the high inhibition toxicity of ILs was mainly related to the length of the single bond of the cation side chain and the surface area of the anion/cation.Subsequently,the molecular docking method was used to reveal the interaction mechanism between ILs and h CES2 at the molecular-protein level.The binding energy indicated that cations bind more strongly to h CES2 than anions.The binding pattern showed that non-bonding interactions such as van der Waals forces between cation side chains and h CES2 contributed positively to the h CES2 inhibition of ILs,which was consistent with the machine learning modeling results.The hydrogen bond was the main force between the anion and h CES2.In summary,our work provided a feasible method for aquatic biological toxicity analysis and human health risk assessment of ILs,which was not only used to explain the toxic mechanism of action of ILs,but also provided a certain theoretical basis for the safe application of ILs.