DCGAT-DTI: dynamic cross-graph attention network for drug-target interaction prediction.

Area of Science:

• Bioinformatics
• Computational Chemistry
• Drug Discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for accelerating drug discovery.
• Current methods often analyze drug-drug and protein-protein similarities independently, failing to capture cross-modal dependencies.
• This limitation hinders the comprehensive modeling of interactions between chemical compounds and proteins.

Purpose of the Study:

• To propose DCGAT-DTI, a deep learning framework designed to improve DTI prediction.
• To dynamically model intra- and cross-graph interactions between drugs and proteins.
• To overcome the limitations of existing methods that process similarity graphs in isolation.

Main Methods:

• Utilizes pretrained language models for initial embedding generation of drugs and proteins.
• Constructs similarity graphs from these embeddings.
• Employs a novel dynamic cross-graph attention (DCGAT) module, incorporating a Graph Convolutional Network-based Cross-Neighborhood Selection network.
• Dynamically selects cross-modal neighbors to integrate information from both drug and protein modalities via attention mechanisms.

Main Results:

• DCGAT-DTI demonstrates superior performance compared to state-of-the-art methods on four benchmark datasets.
• Achieves significant improvements across both balanced and unbalanced datasets, including challenging cold-start scenarios.
• Shows enhanced prediction accuracy for both drugs and proteins, particularly in unbalanced cold-start conditions.

Conclusions:

• DCGAT-DTI effectively models interdependencies between drug and protein modalities for improved DTI prediction.
• The dynamic cross-graph attention mechanism is key to its enhanced performance.
• The framework offers a promising advancement for accelerating the drug discovery pipeline.