# Toxicological Evaluation of Ionic Liquids: QSAR Approach for Acetylcholinesterase Enzyme Inhibition

## TL;DR

This paper develops a model to predict the toxicity of ionic liquids based on their effect on an important enzyme, using a data-driven approach.

## Contribution

The study introduces a QSAR model that incorporates anion structures, which were previously overlooked, to better predict ionic liquid toxicity.

## Key findings

- A QSAR model using 11 COSMO-RS descriptors achieved strong predictive performance (R² = 0.75, RMSE = 0.35).
- The model accounts for anion structures, improving upon previous models that neglected this aspect.
- Predicted toxicity values for new ionic liquids were provided, enhancing safety assessments.

## Abstract

A “quantitative
structure–activity relationship”
(QSAR) model is developed to predict the toxicity of ionic liquids
(ILs) based on the effect on the acetylcholinesterase (AChE) enzyme.
A data set of 243 ILs was compiled and randomly divided into training
(183 ILs) and test (60 ILs) sets to enable both internal and external
validations. To optimize the model performance, a breaking point analysis
was performed to identify the most relevant molecular descriptors.
The analysis revealed that a set of 11 COSMO-RS quantum chemical descriptors
provided near-optimal predictive power, with additional descriptors
offering minimal improvement. A multiple linear regression (MLR) model
was developed by using these descriptors, incorporating both cationic
and anionic molecular features. Internal validation using Leave-One-Out
and Leave-Many-Out cross-validation (Q
2
LOO = 0.79, Q
2
LMO = 0.78) as well as Y-scrambling confirmed the robustness of the
model. External validation on the test set yielded acceptable R
2 = 0.75 and low RMSE = 0.35 values, indicating
strong predictive performance. The developed model outperformed previous
models, particularly by accounting for the influence of anion structures,
which have been largely neglected in earlier works. The final MLR-QSAR
model not only demonstrated statistical reliability but also provided
mechanistic insights into the structural contributions of both ionic
components to IL’s toxicity. Predicted toxicity values (Log
1/EC50) for novel ILs are also presented, expanding our understanding
of IL safety profiles.