HGDTI: predicting drug-target interaction by using information aggregation based on heterogeneous graph neural...

Area of Science:

• Bioinformatics
• Computational Drug Discovery
• Machine Learning

Background:

• Traditional drug discovery involves lengthy wet experiments.
• In silico prediction of drug-target interactions (DTI) can accelerate candidate medication screening.
• Bioinformatics networks offer potential for revealing drug-target connections.

Purpose of the Study:

• To develop an advanced computational model for predicting drug-target interactions (DTI).
• To improve the efficiency and accuracy of drug discovery pipelines.
• To leverage heterogeneous information for enhanced DTI prediction.

Main Methods:

• Developed HGDTI, a heterogeneous graph neural network model.
• Utilized molecular fingerprints and pseudo amino acid composition for feature extraction.
• Employed Bi-LSTM and attention mechanisms for node embedding and neighbor aggregation.
• Incorporated negative sampling to optimize prediction performance.

Main Results:

• HGDTI significantly outperformed existing state-of-the-art DTI prediction models.
• Demonstrated the robustness and rationality of the HGDTI model through rigorous testing.
• Confirmed HGDTI's ability to capture drug and target embeddings using heterogeneous information.

Conclusions:

• HGDTI effectively utilizes heterogeneous information for drug-target embedding and prediction.
• The model provides valuable assistance for accelerating drug development.
• A user-friendly web server for HGDTI is available at http://bioinfo.jcu.edu.cn/hgdti.