Related Experiment Videos
CMAF-DDI: A Knowledge-Enhanced Cross-Modal Fusion Method Leveraging Protein Representation for Multi-Class Drug-Drug
IEEE Journal of Biomedical and Health Informatics
|August 10, 2026
Summary
Predicting drug-drug interactions (DDIs) is vital for patient safety. Our novel CMAF-DDI framework effectively integrates protein sequence, molecular graph, and knowledge graph features for improved DDI prediction.
Area of Science:
- Pharmacology and Bioinformatics
- Computational Drug Discovery
- Artificial Intelligence in Medicine
Background:
- Accurate drug-drug interaction (DDI) prediction is essential for medication safety and personalized medicine.
- Current DDI prediction methods often underutilize target protein sequence information, relying heavily on molecular graphs or knowledge graphs.
- There is a need for advanced computational frameworks that can integrate diverse data sources for more robust DDI prediction.
Purpose of the Study:
- To propose and evaluate CMAF-DDI, a novel multi-class DDI prediction framework.
- To investigate the impact of integrating protein sequence features alongside molecular and knowledge graph features for DDI prediction.
- To develop a bi-level cross-modal fusion module for enhanced DDI prediction.
Main Methods:
- Developed CMAF-DDI, a framework integrating protein sequence, molecular graph, and knowledge graph features.
- Implemented a bi-level cross-modal fusion module comprising Attention Fusion (AF) and Triple-feature Product Fusion (TPF) levels.
- Utilized multi-head attention for global modality dependencies and feature projection into a shared latent space for high-order co-activation.
Main Results:
- CMAF-DDI demonstrated improved multi-class DDI prediction performance on DrugBank and DRKG datasets.
- The proposed framework outperformed existing representative graph-based and multi-source fusion baselines.
- Ablation studies confirmed the significant contribution of protein-enhanced cross-modal fusion to prediction accuracy.
Conclusions:
- The CMAF-DDI framework effectively leverages multi-modal data, particularly protein sequence information, for superior DDI prediction.
- Integrating diverse features through advanced fusion techniques enhances the accuracy and reliability of computational DDI prediction.
- This approach offers a promising direction for improving medication safety and advancing personalized treatment strategies.