MSI-DTI: predicting drug-target interaction based on multi-source information and multi-head self-attention.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate identification of drug-target interactions (DTIs) is vital for drug discovery, virtual screening, and repurposing.
• Existing DTI prediction methods often struggle with data sparsity and the cold-start problem due to reliance on single feature types.

Purpose of the Study:

• To develop a novel framework, MSI-DTI (Multi-Source Information-based Drug-Target Interaction Prediction), to improve the accuracy and robustness of DTI predictions.
• To address the limitations of existing methods by integrating diverse data sources and advanced feature representation techniques.

Main Methods:

• Constructed a comprehensive Drug-Target Knowledge Graph (DTKG).
• Extracted multi-source feature representations from SMILES and amino acid sequences, integrating biometric features and knowledge graph embeddings.
• Employed a multi-head self-attention mechanism with residual connections for effective fusion and capturing higher-order interactions.

Main Results:

• MSI-DTI demonstrated superior performance compared to state-of-the-art methods on the DTKG and two benchmark datasets.
• The framework achieved more accurate and robust predictions for drug-target interactions.
• The integration of multi-source information and advanced attention mechanisms proved effective.

Conclusions:

• MSI-DTI offers a significant advancement in DTI prediction by leveraging multi-source information fusion.
• The proposed framework effectively overcomes the sparsity and cold-start challenges inherent in traditional methods.
• MSI-DTI provides a valuable tool for accelerating drug discovery and development processes.