Interpretable machine learning models for predicting acetylcholinesterase inhibition by ionic liquids.

The widespread application of ionic liquids has been limited by uncertainties related to their ecotoxicity. In this article, we investigate the relationship between molecular descriptors of ionic liquids and acetylcholinesterase (AChE) inhibition using a QSAR approach coupled with machine learning models. An initial set of descriptors underwent a preprocessing and selection process based on multiple criteria, including importance by models, correlation with the target variable, statistical relevance, stability, and non-redundancy, followed by iterative sequential selection. Predictive models were built and evaluated from multiple random partitions of the data, showing high performance (R2>0.8700). The analysis of the descriptors indicated that inhibitory activity is strongly associated with the structural characteristics of the cations, especially properties related to hydrophobicity, electronic distribution, and molecular topology. Cations with aromatic systems showed greater affinity for the enzyme, while very long alkyl chains reduced activity due to steric effects. On the other hand, anions played a secondary role. The results contribute to the understanding of the structure-activity relationship and aid in the development of ionic liquids with lower toxic potential.