QuantumTox: Utilizing quantum chemistry with ensemble learning for molecular toxicity prediction.

Area of Science:

• Computational chemistry
• Drug discovery
• Toxicology

Background:

• Molecular toxicity prediction is crucial for drug discovery and human health.
• Current machine learning models often overlook 3D molecular information.
• Quantum chemical information, including stereostructural details, impacts molecular toxicity.

Purpose of the Study:

• To introduce QuantumTox, the first application of quantum chemistry for drug molecule toxicity prediction.
• To leverage 3D molecular features derived from quantum chemistry.
• To improve the accuracy and generalization of toxicity prediction models.

Main Methods:

• Extracting quantum chemical information as 3D molecular features.
• Utilizing Gradient Boosting Decision Tree and Bagging ensemble learning methods.
• Developing QuantumTox for molecular toxicity assessment.

Main Results:

• QuantumTox consistently outperforms baseline models in toxicity prediction tasks.
• The model demonstrates strong performance even on small datasets (<300 molecules).
• Integration of quantum chemical features significantly enhances predictive accuracy.

Conclusions:

• QuantumTox represents a novel approach by incorporating quantum chemistry into toxicity prediction.
• The method offers improved accuracy and generalization for identifying toxic molecules.
• This approach is valuable for early-stage drug discovery, especially with limited data.