Predictive modeling of antibacterial activity of ionic liquids by machine learning methods

Structural variation and different bioactivity of ionic liquids (ILs) make them highly promising for the development of novel biocides. Application of computational methods to the evaluation of potential antibacterial activity of chemical compounds is a useful, time-and cost-saving tool replacing numerous experimental syntheses. In the present study, quantitative structure-activity relationship (QSAR) modeling is applied to develop models (based on more than 800 data points) aiming to predict the minimal inhibitory concentration (MIC) of ILs against three types of human pathogens - Staphylococcus aureus, Escherichia coli and Pseudomonas aeruginosa. The random forest model with the AlvaDesc descriptors in general demonstrates the best performance for all the three types of bacteria and is suggested as a final model. To interpret the final model and determine the most sig-nificant descriptors, a SHapley Additive exPlanation (SHAP) method was applied. Six amino acid ILs, which were synthesized for the first time, and five halogenide ionic liquids purchased, all based on 1-alkyl-3methylimidozo-lium cations with different alkyl chain lengths, C-10, C-12 and C-14, are tested in vitro and used to validate the developed QSAR models. The data sets and developed model are available free of charge at http://ochem. eu/article/147386.

Automated summary

In this study, a quantitative structure-activity relationship (QSAR) model was developed to predict the minimal inhibitory concentration (MIC) of ionic liquids (ILs) against three human pathogens. The random forest model with the AlvaDesc descriptors demonstrated the best performance. Six amino acid ILs were synthesized for the first time and five halogenide ILs were purchased for validation.