A Biological Feature and Heterogeneous Network Representation Learning-Based Framework for Drug-Target Interaction...

Area of Science:

• Computational Biology
• Pharmacology
• Bioinformatics

Background:

• Drug-target interaction (DTI) prediction is vital for efficient drug discovery.
• Traditional experimental methods for DTI determination are costly, time-consuming, and labor-intensive.
• Developing computational approaches is essential to accelerate the drug discovery pipeline.

Purpose of the Study:

• To introduce BG-DTI, a novel learning-based framework for predicting drug-target interactions.
• To integrate biological sequence features with heterogeneous network information for enhanced DTI prediction.
• To provide a more efficient and accurate alternative to experimental DTI determination.

Main Methods:

• Encoding drugs and targets using sequence-derived biological features.
• Constructing drug-drug and target-target similarity networks to capture biological relationships.
• Employing graph convolutional networks (GCN) and graph attention networks (GAT) for feature representation learning.
• Utilizing a random forest classifier on fused descriptors for final DTI prediction.

Main Results:

• BG-DTI achieved a high average AUC of 0.938 and AUPR of 0.930.
• The proposed framework outperformed five existing state-of-the-art DTI prediction methods.
• Demonstrated the efficacy of combining sequence features and network information for DTI prediction.

Conclusions:

• BG-DTI offers a powerful and accurate computational tool for predicting drug-target interactions.
• The framework has the potential to significantly facilitate drug discovery and drug repurposing initiatives.
• The integration of graph representation learning with biological features represents a promising direction in DTI prediction research.