QSAR Classification Modeling Using Machine Learning with a Consensus-Based Approach for Multivariate Chemical Hazard...

Area of Science:

• Computational toxicology
• Cheminformatics
• Machine learning in drug discovery

Background:

• Traditional toxicity testing is time-consuming and ethically challenging.
• Predictive toxicology models are crucial for efficient chemical safety assessment.
• In silico methods offer a promising alternative to in vivo testing.

Purpose of the Study:

• To develop and validate a hybrid machine learning model for predicting chemical toxicity across eight critical endpoints.
• To leverage advanced cheminformatics tools for feature extraction and model development.
• To establish a robust computational framework for in silico chemical safety assessment.

Main Methods:

• Utilized hybrid machine learning models, including Random Forest, XGBoost, and SVM classifiers.
• Extracted molecular descriptors such as Morgan fingerprints, MACCS keys, and physicochemical properties.
• Developed a consensus model by selecting optimal classifiers for each descriptor and endpoint.

Main Results:

• Achieved strong predictive performance with Area Under the Curve (AUC) scores ranging from 0.78 to 0.90 across eight toxicity endpoints.
• Demonstrated the robustness and accuracy of the developed in silico toxicity prediction framework.
• Validated the model's effectiveness in predicting cardiac, inhalation, dermal, oral, skin irritation, skin sensitization, eye irritation, and respiratory irritation.

Conclusions:

• The developed hybrid machine learning framework provides a reliable and ethical in silico approach for toxicity prediction.
• This computational method supports regulatory and research applications in chemical safety assessment.
• Highlights the potential of advanced computational techniques to advance toxicological evaluations.