VHGAE: Drug-Target Interaction Prediction Model Based on Heterogeneous Graph Variational Autoencoder

Area of Science:

• Computational biology
• Bioinformatics
• Network science

Background:

• Drug-target interaction (DTI) identification is essential for drug discovery and repositioning.
• Traditional biological methods for DTI identification are time-consuming and expensive.
• Heterogeneous network methods offer a faster approach to DTI prediction, but sparsity in known DTI data poses a challenge for graph convolutional networks.

Purpose of the Study:

• To propose VHGAE, a novel method based on a heterogeneous graph variational autoencoder for accurate drug-target interaction prediction.
• To address the data sparsity issue in heterogeneous networks for DTI prediction.
• To leverage multi-source prior knowledge for enhanced DTI prediction.

Main Methods:

• Constructed a heterogeneous network by integrating diverse prior knowledge about drugs and targets.
• Applied the weighted k-nearest neighbor algorithm to densify the drug-target interaction network, enhancing node connectivity.
• Utilized a weighted graph convolutional network within a variational graph autoencoder framework to strengthen edge weights.
• Incorporated a variational expectation maximization algorithm to recover potential relationships within the sparse network.

Main Results:

• The proposed VHGAE method demonstrated superior performance in predicting drug-target interactions.
• VHGAE outperformed nine existing state-of-the-art DTI prediction methods on two benchmark datasets.
• The results highlight the effectiveness of VHGAE's approach to multi-source data fusion and sparse network processing.

Conclusions:

• VHGAE significantly improves drug-target interaction prediction accuracy by effectively handling sparse heterogeneous networks.
• The method's ability to integrate multi-source data and process sparse networks is key to its enhanced performance.
• VHGAE offers a promising computational approach to accelerate drug discovery and repositioning efforts.