Hierarchical multimodal self-attention-based graph neural network for DTI prediction.

Area of Science:

• Bioinformatics
• Computational Chemistry
• Machine Learning

Background:

• Drug-target interactions (DTIs) are vital for efficient drug development.
• Current deep learning models often rely on single data representations, limiting comprehensive feature analysis.
• Existing multimodal approaches fail to simultaneously capture intra- and inter-modal interactions, hindering DTI prediction accuracy.

Purpose of the Study:

• To propose a novel deep learning model for accurate and efficient drug-target interaction prediction.
• To address the limitations of single-modal representations and incomplete multimodal fusion in existing DTI prediction models.
• To enhance the feature representation capabilities by simultaneously considering intra- and inter-modal interactions.

Main Methods:

• Developed a hierarchical multimodal self-attention-based graph neural network (HMSA-DTI).
• Input data includes drug SMILES, molecular graphs, protein sequences, and 2-mer sequences.
• Employed a hierarchical multimodal self-attention mechanism for deep fusion of drug and protein features, capturing both intra- and inter-modal interactions.

Main Results:

• The proposed HMSA-DTI model demonstrated superior performance compared to baseline methods.
• Achieved significant advantages across multiple evaluation metrics.
• Validated on five benchmark datasets, confirming its effectiveness in DTI prediction.

Conclusions:

• HMSA-DTI effectively integrates multimodal data for enhanced DTI prediction.
• The hierarchical multimodal self-attention mechanism captures crucial intra- and inter-modal interactions.
• This approach offers a promising direction for improving drug development efficiency through accurate DTI prediction.