DSANIB: Drug-Target Interaction Predictions With Dual-View Synergistic Attention Network and Information Bottleneck...

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Drug repositioning relies on accurate drug-target interaction (DTI) prediction.
• Experimental DTI identification is costly and time-consuming.
• Deep learning models show promise but struggle with capturing local interactions and filtering redundant features.

Purpose of the Study:

• To develop a novel deep learning approach, DSANIB, for predicting drug-target interactions.
• To address challenges in capturing higher-order substructure embeddings and obtaining discriminative representations.

Main Methods:

• DSANIB utilizes a Drug-Target Specific Attention Network (DSAN) component with Inter-view and Intra-view Attention Modules.
• The Information Bottleneck (IB) strategy is employed to retain relevant information and minimize redundancy.
• The model learns higher-order substructure embeddings for drugs and targets.

Main Results:

• DSANIB demonstrated superior performance compared to state-of-the-art (SOTA) prediction models.
• The approach effectively captures local drug-target interactions and learns discriminative embeddings.
• Visualization of learned embeddings provided interpretable insights into prediction outcomes.

Conclusions:

• DSANIB offers an effective deep learning framework for DTI prediction.
• The model's ability to learn discriminative representations enhances drug repositioning strategies.
• The approach provides interpretable insights, aiding in understanding prediction mechanisms.