GADTI: Graph Autoencoder Approach for DTI Prediction From Heterogeneous Network

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Drug-target interaction (DTI) identification is crucial for drug development.
• Experimental methods for DTI discovery are costly and have limited coverage.
• Existing computational methods for DTI prediction require accuracy improvements.

Purpose of the Study:

• To propose a novel graph autoencoder approach for drug-target interaction prediction (GADTI).
• To enhance the accuracy and efficiency of predicting potential DTIs.
• To leverage diverse biological datasets for improved DTI prediction.

Main Methods:

• Developed GADTI, a graph autoencoder model integrating diverse drug and target datasets into a heterogeneous network.
• Employed a graph convolutional network (GCN) and random walk with restart (RWR) in the encoder for enhanced node information.
• Utilized DistMult, a matrix factorization model, in the decoder for DTI prediction.

Main Results:

• GADTI demonstrated superior performance over baseline methods in 10-fold cross-validation experiments.
• The model achieved higher accuracy as measured by the area under the receiver operator characteristic curve (AUC) and area under the precision-recall curve (AUPRC).
• A case study showed GADTI successfully predicted 54.8% of newly approved DTIs from the Drugbank dataset (V5.1.8).

Conclusions:

• GADTI offers a powerful and accurate computational approach for predicting drug-target interactions.
• The integrated GCN and RWR methods effectively capture complex relationships within biological networks.
• GADTI shows significant potential for accelerating drug discovery and development pipelines.