GGL-Tox: Geometric Graph Learning for Toxicity Prediction.

Area of Science:

• Computational chemistry
• Toxicology
• Machine learning

Background:

• Toxicity analysis is crucial in drug design and discovery.
• Machine learning (ML) offers accurate, efficient, and cost-effective solutions.
• The US Toxicology in the 21st Century (Tox21) initiative screened numerous compounds for toxic effects.

Purpose of the Study:

• To develop a novel computational model for toxicity prediction.
• To leverage multiscale weighted colored graph (MWCG) theory for toxicity analysis.
• To evaluate the model's performance on the Tox21 Data Challenge benchmarks.

Main Methods:

• Developed a Geometric Graph Learning Toxicity (GGL-Tox) model.
• Integrated MWCG features with the Gradient Boosting Decision Tree (GBDT) algorithm.
• Utilized benchmark datasets from the Tox21 Data Challenge for validation.

Main Results:

• The GGL-Tox model demonstrated significant utility and usefulness in toxicity analysis.
• Performance was evaluated against state-of-the-art models.
• The model proved to be accurate and efficient for predicting compound toxicity.

Conclusions:

• The GGL-Tox model, integrating MWCG features and GBDT, is a powerful tool for toxicity prediction.
• This approach advances ML applications in drug discovery and toxicology.
• GGL-Tox offers a promising alternative for efficient and accurate toxicity assessment.