Graph-DTI: A New Model for Drug-target Interaction Prediction Based on Heterogenous Network Graph Embedding

Area of Science:

• Computational chemistry and cheminformatics
• Bioinformatics and systems biology
• Machine learning and artificial intelligence in drug discovery

Background:

• Accurate drug-target interaction (DTI) prediction is crucial for guiding drug discovery and development.
• Existing machine learning methods often struggle with integrating diverse data sources and capturing complex relationships between drugs and protein targets.
• Previous studies using heterogeneous network graphs for DTI prediction have limitations in representing neighborhood information.

Purpose of the Study:

• To develop an end-to-end learning model, Graph-Drug-Target Interaction (Graph-DTI), for predicting DTIs.
• To integrate various data types within a heterogeneous network, including drug-drug interactions, protein-protein interactions, drug structure similarity, and protein sequence similarity.
• To explore automatic learning of topology-maintaining representations for drugs and targets to enhance DTI prediction accuracy.

Main Methods:

• Construction of a heterogeneous network integrating drugs, targets, and their associated interaction and similarity data (DrugBank, HPRD, RDKit, Smith-Waterman).
• Application of a graph neural network-inspired graph auto-encoding method to extract high-order structural information and node representations.
• Prediction of potential DTIs using the learned representations, followed by secondary classification of the obtained samples.

Main Results:

• The Graph-DTI model demonstrated superior performance compared to all baseline methods.
• Performance was evaluated using the area under the precision-recall curve (AUPR) and the area under the receiver operating characteristic curve (AUC).
• Graph-DTI achieved better results in both AUPR and AUC metrics, indicating enhanced prediction accuracy.

Conclusions:

• Graph-DTI significantly outperforms existing DTI prediction methods, offering improved prediction performance.
• The model effectively classifies drugs based on their targets and vice versa.
• Graph-DTI serves as a powerful and more effective tool for drug research, development, and repositioning compared to previous approaches.