Machine learning-assisted comparative QSTR, i-QSTTR, qRASTR, and i-qRASTTR modelling for toxicity of Ionic liquids against three different bacteria S. aureus, E. coli, and P. aeruginosa.

Ionic liquids (ILs) with tunable structures have emerged as promising next-generation biocides. In this study, we presented an ML-based q-RASTR framework, along with i-qRASTTR approach, to predict the toxicity of ILs against different bacteria. Various ML algorithms were employed for constructing several predictive toxicity models. Comparative performance analysis revealed that the q-RASTR (Q: 0.878, Q: 0.878, MAE: 0.355) and i-qRASTTR models (Q: 0.908- 0.917, Q: 0.908-0.917, MAE: 0.292-0.187) performed better than traditional QSTR and i-QSTTR models in terms of external predictivity. The i-qRASTTR models also revealed strong interspecies correlations, particularly between S. aureus and E. coli, and between S. aureus and P. aeruginosa. Toxicity models revealed that molecular features such as cation size, branching, and length of the alkyl chain significantly enhance the toxicity of ILs, whereas the presence of oxygen-containing side chains tends to reduce their toxicity in gram +ve bacteria. Moreover, interspecies correlation analysis indicated shared toxicity mechanisms across gram +ve and gram -ve bacteria, with larger and more hydrophobic cations demonstrating enhanced membrane disruption in both cases. Overall, our findings highlighted that the integration of data-driven qRASTR and i-qRASTTR modelling offers a powerful strategy to understand structure-toxicity and toxicity-toxicity relationships of ILs.