Related Experiment Video
Updated: May 25, 2026
10:10
Assessment of the Acute Inhalation Toxicity of Airborne Particles by Exposing Cultivated Human Lung Cells at the Air-Liquid Interface
Published on: February 23, 2020
Machine learning modeling of acute inhalation toxicity using an RFA-RFR framework, supported by explainable AI and
Vijay H Masand1, M M Rathore1, Jayant R Bansod1
1Department of Chemistry, Vidya Bharati Mahavidyalaya, Amravati 444 602, Maharashtra, India.
Journal of Pharmacological and Toxicological Methods
|May 23, 2026
Summary
This study developed machine learning models to predict acute inhalation toxicity (pLC₅₀) for 552 molecules. The models achieved high accuracy, offering improved predictions and mechanistic insights for chemical safety assessments.
Area of Science:
- Computational chemistry
- Toxicology
- Machine learning
Background:
- Accurate prediction of acute inhalation toxicity is crucial for chemical safety.
- Existing quantitative structure-activity relationship (QSAR) models have limitations in predictive accuracy and applicability.
- Developing robust and interpretable models is essential for understanding chemical hazards.
Purpose of the Study:
- To develop and validate machine learning models for predicting the acute inhalation toxicity (pLC₅₀) of diverse molecules.
- To identify key molecular descriptors influencing inhalation toxicity.
- To provide a more accurate and broadly applicable predictive tool compared to existing QSAR models.
Main Methods:
- Dataset curation from the Integrated Chemical Environment (ICE) database for 552 molecules.
- Generation of 3D molecular structures and calculation of ~58,000 molecular descriptors.
- Feature selection using the Red Fox Algorithm (RFA) and model development with Random Forest Regressor (RFR).
- Model interpretation using Shapley additive explanations (SHAP) values.
Main Results:
- The RFA-RFR model achieved robust statistical performance with R²tr=0.907, R²L10%O=0.902, and Q₂F₃=0.887.
- Model validation confirmed robustness through leave-10%-out cross-validation and external set validation.
- Benchmarking demonstrated improved predictive accuracy and broader chemical applicability compared to previous QSAR models.
- Mechanistic interpretation identified molecular weight, hydrogen bond acceptors, lipophilicity, and specific carbon-oxygen arrangements as key toxicity drivers.
Conclusions:
- The developed machine learning approach, integrating feature optimization and interpretable models, provides accurate predictions of acute inhalation toxicity.
- The models offer valuable mechanistic insights into the factors governing inhalation toxicity.
- This study underscores the utility of interpretable machine learning for advancing chemical safety assessments.
