VGAEDTI: drug-target interaction prediction based on variational inference and graph autoencoder

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Accurate identification of Drug-Target Interactions (DTIs) is vital for drug development and repurposing.
• Existing methods often fail to leverage multi-source data or capture complex inter-source relationships.
• Challenges remain in mining high-dimensional drug and target data for improved model accuracy and robustness.

Purpose of the Study:

• To address limitations in current DTI prediction methods.
• To develop a novel model for accurate and robust DTI prediction.
• To leverage multi-source data and advanced feature extraction techniques.

Main Methods:

• Proposed a novel prediction model named VGAEDTI.
• Constructed a heterogeneous network integrating multiple drug and target data sources.
• Employed a variational graph autoencoder (VGAE) for feature representation learning and a graph autoencoder (GAE) for label propagation.

Main Results:

• VGAEDTI demonstrated superior prediction accuracy compared to six existing DTI prediction methods on two public datasets.
• The model effectively infers feature representations from drug and target spaces.
• Experimental results validate the model's capability in predicting novel DTIs.

Conclusions:

• VGAEDTI offers an effective computational tool for accelerating drug discovery and repurposing efforts.
• The model's performance highlights the benefits of integrating multi-source data and advanced autoencoder architectures.
• This approach provides a robust framework for understanding complex drug-target relationships.