Predicting drug-drug interactions based on multi-view and multichannel attention deep learning.

Area of Science:

• Pharmacology
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Drug-drug interactions (DDIs) are critical in drug research for understanding drug function and developing new therapeutics.
• Current DDI prediction methods often fail to fully leverage multi-source drug data, limiting their effectiveness.
• Biomedical knowledge graphs (KGs) and drug attributes are commonly used but can be integrated more comprehensively.

Purpose of the Study:

• To propose a novel multi-view and multichannel attention deep learning (MMADL) model for enhanced DDI prediction.
• To effectively extract rich drug features from multi-source databases, including drug attributes and related entities.
• To improve the accuracy and effectiveness of DDI prediction by considering feature consistency and complementarity.

Main Methods:

• Developed a multi-view and multichannel attention deep learning (MMADL) model.
• Employed a single-layer perceptron encoder to create multi-view drug representation vectors.
• Utilized a multichannel attention mechanism to learn the importance of drug features for DDI prediction.

Main Results:

• The MMADL model achieved an accuracy of 93.05% and a precision-recall curve of 95.94%.
• The model demonstrated superior performance compared to existing state-of-the-art methods.
• MMADL successfully integrated multi-source drug information for robust DDI prediction.

Conclusions:

• The proposed MMADL model offers a significant advancement in DDI prediction accuracy.
• Integrating multi-source drug data through a multi-view and multichannel attention approach is highly effective.
• This method holds promise for accelerating the development of safer and more effective therapeutic drugs.