FGMA: A Functional Group-enhanced Deep Learning Method for Drug Toxicity Prediction.

Area of Science:

• Computational chemistry
• Drug discovery
• Toxicology

Background:

• Accurate drug toxicity prediction is crucial for efficient drug discovery.
• Existing atom-based methods face computational challenges with complex molecules.
• There is a need for more efficient and powerful predictive models.

Purpose of the Study:

• Introduce FGMA, a novel framework for efficient drug toxicity prediction.
• Address the computational inefficiency of current methods.
• Enhance the accuracy and interpretability of toxicity assessments.

Main Methods:

• FGMA integrates functional groups with molecular fingerprints for multi-view encoding.
• Key functional groups are extracted as unified entities to reduce computational load.
• A cross-attention mechanism combines functional groups with MACCS fingerprints for comprehensive molecular representation.

Main Results:

• FGMA demonstrates high accuracy and robustness in toxicity prediction tasks.
• Interpretability analyses successfully identified specific structural features and functional groups linked to toxicity.
• The framework effectively balances computational efficiency with representational power.

Conclusions:

• Combining functional groups and fingerprints is an effective strategy for toxicity prediction.
• FGMA provides valuable insights into structure-toxicity relationships for safer drug design.
• FGMA accelerates drug discovery through faster, more insightful safety assessments.