Predicting Drug-Protein Interactions by Self-Adaptively Adjusting the Topological Structure of the Heterogeneous...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Predicting drug-protein interactions (DPIs) is crucial for drug discovery and repurposing.
• Existing methods struggle with incomplete data and unreliable heterogeneous networks due to unknown drug/protein functions.
• Graph neural networks (GNNs) offer powerful tools but require robust network structures.

Purpose of the Study:

• To develop a novel DPI prediction method that can self-adaptively adjust the topological structure of heterogeneous networks.
• To address challenges posed by incomplete data and unreliable networks in DPI prediction.
• To improve the accuracy and reliability of computational drug discovery pipelines.

Main Methods:

• Proposed SATS, a method utilizing a graph attention network (GAN) for representation learning within drug-protein heterogeneous networks.
• Implemented self-adaptive learning of node relationships based on attributes.
• Incorporated dynamic adjustment of network topology based on model training loss.
• Predicted drug-protein interaction propensity using learned embeddings.

Main Results:

• SATS effectively improved the topological structure of the heterogeneous network.
• Demonstrated superior performance compared to state-of-the-art DPI prediction methods across various metrics.
• Showcased utility in handling incomplete data and unreliable networks.
• Case studies confirmed SATS's capability in discovering novel drug-protein interactions.

Conclusions:

• SATS offers a robust solution for predicting drug-protein interactions, particularly in scenarios with data limitations.
• The adaptive network structure adjustment mechanism enhances prediction accuracy and reliability.
• SATS holds significant potential for accelerating drug discovery and repurposing efforts by identifying novel interactions.