A Node-Adaptive Feature Fusion Network for Drug-Target Interaction Prediction Based on Multi-View Graphs.

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Existing drug-target interaction (DTI) prediction methods struggle with sparse data, complex relationships, and imbalanced information.
• Accurate DTI prediction is crucial for drug discovery and repurposing.

Purpose of the Study:

• To develop a novel network, NAFF-DTI, for improved DTI prediction.
• To address challenges of sparse data and heterogeneous information in DTI prediction.

Main Methods:

• Proposed NAFF-DTI, a node-level adaptive feature fusion network based on multi-view graphs.
• Uniformly represented drug similarity, target similarity, and known DTIs as multiple relational views.
• Employed graph encoding, cross-view representation learning, cross-view feature discrepancy modeling, and adaptive fusion.

Main Results:

• NAFF-DTI achieved superior performance (AUC and AUPR) across five benchmark datasets.
• Demonstrated average relative improvements of 3.81% in AUC and 3.23% in AUPR over strongest baselines.
• Showcased enhanced utilization of multi-source information and stable predictions across data distributions.

Conclusions:

• NAFF-DTI offers a robust computational framework for DTI candidate prioritization.
• The model effectively generates hypotheses for drug repurposing.
• NAFF-DTI advances the field of computational drug discovery.