MolFPG: Multi-level fingerprint-based Graph Transformer for accurate and robust drug toxicity prediction

Area of Science:

• Computational chemistry
• Drug discovery
• Toxicology

Background:

• Drug toxicity prediction is crucial for efficient drug development, reducing risks and costs.
• Existing methods like handcrafted features and molecular graphs have limitations in molecular representation learning.

Purpose of the Study:

• To develop an innovative framework for interpretable drug toxicity prediction.
• To improve molecular representation learning for enhanced drug safety assessment.

Main Methods:

• Developed the MolFPG (Molecular Fingerprint Graph Transformer) framework.
• Integrated multiple molecular fingerprinting techniques with Graph Transformer-based representation.
• Incorporated a global-aware module for interpretable toxicity prediction.

Main Results:

• Achieved high accuracy and reliability in predicting drug toxicity.
• Demonstrated improved interpretability by exploring drug features and toxicity relationships.
• Highlighted the potential of Graph Transformers and multi-level fingerprints in drug discovery.

Conclusions:

• The MolFPG framework offers a reliable and effective approach for drug toxicity prediction.
• This study provides vital support for advancing drug development and toxicity assessment.
• The findings underscore the value of advanced molecular representations for drug safety.