MUFFIN: multi-scale feature fusion for drug-drug interaction prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Adverse drug-drug interactions (DDIs) are a significant cause of morbidity and mortality, necessitating accurate identification.
• Traditional machine learning models for DDI prediction often lack generalization due to reliance on handcrafted features.
• Existing deep learning methods for DDI prediction primarily focus on molecular structure or sequence data, neglecting crucial relational and topological information within biomedical knowledge graphs.

Purpose of the Study:

• To develop a novel deep learning model that integrates both drug molecular structure and knowledge graph (KG) semantic information for improved DDI prediction.
• To address the limitations of existing models by leveraging multi-modal data sources for more robust drug-drug interaction identification.

Main Methods:

• Proposed MUFFIN, a multi-scale feature fusion deep learning model.
• Developed a bi-level cross strategy (cross- and scalar-level components) to effectively fuse multi-modal features.
• Utilized drug molecular graph information and information from large-scale biomedical KGs to learn joint drug representations.

Main Results:

• MUFFIN demonstrated superior performance compared to state-of-the-art baselines across three datasets and three DDI prediction tasks (binary-class, multi-class, and multi-label).
• The model effectively alleviates the restriction of limited labeled data by integrating information from large-scale KGs and drug molecular graphs.
• Achieved improved accuracy in predicting unknown drug-drug interactions by jointly learning from diverse data modalities.

Conclusions:

• The proposed MUFFIN model offers a powerful approach for DDI prediction by effectively integrating drug molecular structure and KG semantic information.
• This multi-modal fusion strategy enhances the generalization and accuracy of computational models for identifying potential adverse drug-drug interactions.
• MUFFIN represents a significant advancement in computational pharmacology, aiding clinicians and researchers in mitigating DDI-related risks.