A Multi-View Feature-Based Interpretable Deep Learning Framework for Drug-Drug Interaction Prediction.

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in medicine

Background:

• Drug-drug interactions (DDIs) pose significant risks, necessitating accurate prediction methods.
• Current computational DDI prediction models often rely on limited single-view features, hindering performance and interpretability.
• A gap exists in interpretability research for multi-view feature-based DDI prediction.

Purpose of the Study:

• To develop a multi-view feature-based interpretable deep learning framework for enhanced DDI prediction.
• To improve the accuracy and interpretability of computational DDI prediction by integrating diverse molecular features.
• To address the limitations of single-view approaches in current DDI prediction models.

Main Methods:

• Employed a Message Passing Neural Network (MPNN) to extract atomic-view features from molecular graphs.
• Utilized transformer encoders to learn substructure-view embeddings from drug SMILES strings.
• Integrated atomic and substructure features into a holistic drug embedding matrix for a multi-view deep learning framework (MI-DDI).
• Developed an interaction module for interpretable DDI prediction and weight matrix construction.

Main Results:

• MI-DDI demonstrated superior performance over existing benchmarks on the BIOSNAP and DrugBank datasets, with average improvements of 3% and 1%, respectively.
• Experiments confirmed the importance of atomic-view information for DDI prediction accuracy.
• The proposed interaction module effectively learned information crucial for precise and interpretable DDI prediction.

Conclusions:

• MI-DDI offers a significant advancement in DDI prediction by leveraging multi-view features for improved accuracy and interpretability.
• The framework provides a tractable path for understanding drug interactions, crucial for clinical safety.
• The findings highlight the potential of multi-view deep learning in pharmaceutical research and drug safety.