Ionic Liquid Toxicity Prediction And De Novo Design Based On Big Data And Machine Learning

Ionic liquids（ILs）,as a novel solvent,have been widely applied in the fields of chemical synthesis,biomedicine,and catalysis due to their unique physicochemical properties.However,with the expansion of the usage of ILs,more and more studies have shown that ILs pose certain risks to the ecological environment and human health.Therefore,the development of environmentally friendly ILs is urgent.However,ILs have a high degree of structural design freedom,and the enormous time and material costs of experimental research cannot meet production needs.In addition,the scattered and insufficient data on ILs,and the unclear toxicity mechanism,are also important factors restricting the assessment of ILs toxicity and the design of environmentally friendly ILs.Therefore,this project proposes a toxicological prediction and de novo design of ILs based on big data and machine learning,and the specific research content is as follows:（1）The only currently available database of ILs toxicity,ILTox,was constructed,effectively solving the problem of lack of ionic liquid toxicity data.More than 6700 ILs toxicity data are stored in the database,and all ILs structures are in SMILES format,which can be directly input into machine learning models,laying the foundation for in-depth analysis of quantitative relationships between ILs structures and toxicity.To test the usability of the database,this study constructed several prediction models based on four datasets（E.coli,ACh E,rat IPC-81 cells,and Vibrio fischeri）selected from the ILTox database and using molecular fingerprint descriptors and machine learning algorithms,and all models achieved a prediction accuracy of R²>0.61.In addition,feature importance analysis showed that the important molecular features affecting the toxicity of ILs were mainly derived from the head group type of cations,side chain alkyl groups.（2）Based on machine learning and molecular simulation methods,we have analyzed the quantitative relationship between the structure of ILs and their inhibition of acetylcholinesterase（ACh E）activity,and deeply explored the relevant toxicity mechanisms.Our results demonstrate that by integrating multiple machine learning methods,more reliable and stable QSAR models can be obtained（with cross-validation and external validation R²>0.85）.In addition,end-to-end deep learning（such as VGG19＿BN）can directly extract structural features of the interaction between ILs and ACh E,without the need for descriptor calculation.Molecular dynamics simulations show that the cations and organic anions of ILs mainly bind to specific amino acid residues of ACh E through non-covalent interactions such asπinteraction and hydrogen bonding.Calculation of binding free energy indicates that electrostatic interactions(ΔE_ele<-285 k J/mol)are the main driving force for the binding of ILs to ACh E.（3）Furthermore,this study proposes an environmentally friendly framework for the de novo design of ionic liquids based on a GPstack-RNN.The framework consists of a generative model and a predictive model.The generative model,trained on 1,183 ILs experimentally synthesized in ILTox,learns the composition rules of ILs and generates new ILs.The predictive model then screens the generated ILs for desired properties.Using this approach and based on the SMILES representation of ILs,342 structurally reasonable and usable virtual ILs were generated.Through the predictive model,an IL with desired properties（good antibacterial effect and low cell toxicity）was selected.The resulting IL（IL 144）showed the following properties:predicted Hela cell EC ₅₀ value of 634.25 mg/L,experimental value of 611 mg/L;predicted E.coli MIC value of 9.08 mg/L,experimental value of 9.5 mg/L.This provides a feasible method for designing more environmentally friendly ILs in the future.In summary,the ILTox database of ILs toxicity,machine learning prediction models,molecular docking and molecular dynamics simulation mechanism explanation system,and the from-scratch design framework for environmentally friendly ILs proposed in this study provide theoretical support for the biotoxicity and environmental risk assessment of ILs,and lay the foundation for the design of environmentally friendly ILs in the future.