An Explainable Supervised Machine Learning Model for Predicting Respiratory Toxicity of Chemicals Using Optimal...

Area of Science:

• Computational toxicology
• Medicinal chemistry
• Pharmacology

Background:

• Respiratory toxicity poses a significant public health risk, necessitating accurate predictive tools for drug and chemical safety.
• The pharmaceutical and chemical industries require reliable computational methods to assess compound respiratory toxicity early in development.

Purpose of the Study:

• To develop quantitative structure-activity relationship (QSAR) models for predicting respiratory toxicity.
• To identify optimal molecular descriptors and machine learning algorithms for respiratory toxicity assessment.

Main Methods:

• Exploration of various feature selection techniques to identify key molecular descriptors.
• Implementation and comparison of eight distinct machine learning algorithms for model construction.
• Validation using 10-fold cross-validation and evaluation on an independent test set.

Main Results:

• The Support Vector Machine (SVM) classifier demonstrated superior performance compared to other models.
• The best SVM model achieved 86.2% prediction accuracy and a Matthews Correlation Coefficient (MCC) of 0.722 on the test set.
• SHapley Additive exPlanations (SHAP) were used to interpret model predictions and identify critical toxicity-influencing descriptors.

Conclusions:

• The developed SVM model offers a precise and efficient tool for predicting respiratory toxicity.
• This model can significantly benefit early-stage drug discovery by identifying potentially toxic compounds.
• The interpretability of the model enhances understanding of structure-toxicity relationships.