Drug-Target Interaction Prediction via Dual-Interaction Fusion.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Accurate drug-target interaction (DTI) prediction is vital for drug discovery.
• Experimental assays are costly, and current computational models struggle with multi-scale features and cross-modal information fusion.
• Fine-grained drug-protein interaction modeling remains a challenge.

Purpose of the Study:

• To propose Gated-Attention Dual-Fusion Drug-Target Interaction (GADFDTI), a novel computational model for DTI prediction.
• To address limitations in capturing multi-scale features and fusing cross-modal information in existing DTI models.
• To develop an efficient in silico prescreening tool for drug lead discovery.

Main Methods:

• Developed a fusion module constructing an atom-residue similarity field, refined by a 2D neighborhood operator and gated bidirectional aggregation.
• Integrated a multi-scale dense GCN for drug graphs and a masked multi-scale self-attention protein encoder with a 1D-CNN branch.
• Evaluated GADFDTI on Human and *C. elegans* benchmarks.

Main Results:

• GADFDTI achieved high AUC values of 0.986 on Human and 0.996 on *C. elegans*, outperforming recent DTI models.
• The model demonstrated significant improvements in precision and recall.
• A SARS-CoV-2 case study showed GADFDTI's ability to prioritize clinically supported antiviral agents.

Conclusions:

• GADFDTI effectively models fine-grained drug-protein interactions by integrating multi-scale features and cross-modal information.
• The proposed model offers superior performance compared to existing DTI prediction methods.
• GADFDTI shows potential as an efficient in silico prescreening tool for identifying drug leads.