Interactive multi-hypergraph inferring and channel-enhanced and attribute-enhanced learning for drug-related side...

Area of Science:

• Pharmacology
• Computational Biology
• Bioinformatics

Background:

• Identifying drug side effects is crucial for patient safety and successful drug development.
• Existing methods often fail to fully capture complex biological associations between drugs and side effects.
• Functional similarity among drugs frequently leads to shared side effect profiles.

Purpose of the Study:

• To develop an advanced prediction model for identifying potential drug side effects.
• To improve the mining of complex associations between drugs and their adverse effects.
• To enhance drug safety and reduce drug development risks through accurate side effect prediction.

Main Methods:

• Proposed an interactive multi-hypergraph inferring and channel-enhanced, attribute-enhanced learning model (ICAL).
• Designed a hypergraph architecture to represent complex correlations and global relationships between drugs and side effects.
• Utilized an interactive hypergraph neural network with channel-level and attribute-level attention mechanisms for feature learning.

Main Results:

• ICAL outperformed seven advanced prediction methods in AUC, AUPR, and recall rates.
• Ablation studies confirmed the effectiveness of global correlation learning and enhanced pairwise attribute learning.
• Case studies demonstrated ICAL's capability in discovering reliable candidate side effects for drugs.

Conclusions:

• The ICAL model offers a significant advancement in predicting drug side effects.
• This approach effectively integrates biological features and complex relationships for improved accuracy.
• ICAL holds promise for enhancing drug safety and streamlining the drug development pipeline.