SGFNNs: Signed Graph Filtering-based Neural Networks for Predicting Drug-Drug Interactions.

Area of Science:

• Computational chemistry
• Pharmacology
• Machine learning

Background:

• Drug-drug interactions (DDIs) are critical for public health and drug development.
• Graph Neural Networks (GNNs) are increasingly used in drug discovery for link prediction.
• Existing GNN models often use unsigned graphs, failing to capture disassortative drug relationships.

Purpose of the Study:

• To develop a novel Signed Graph Neural Network (Signed GNN) framework for DDI prediction.
• To model both assortative and disassortative relationships between drug pairs.
• To improve the accuracy and comprehensiveness of DDI prediction.

Main Methods:

• Proposed Signed GNNs to represent assortative and disassortative drug relationships.
• Divided signed graphs into two unsigned subgraphs for spectral filtering.
• Developed Signed Graph Filtering-based Neural Networks (SGFNNs) integrating graph structures and node attributes.
• Employed an end-to-end framework with SGFNNs and a discriminator for joint training.

Main Results:

• The proposed framework achieved significant improvements in DDI prediction accuracy over baseline methods.
• Experimental results on two prediction tasks demonstrated the effectiveness of the Signed GNN approach.
• A case study validated the practical utility and accuracy of the developed method.

Conclusions:

• Signed GNNs effectively capture complex drug-drug relationships, including negative correlations.
• The proposed end-to-end framework enhances DDI prediction accuracy.
• This approach offers a promising direction for drug discovery and safety assessment.