IHDFN-DTI: Interpretable Hybrid Deep Feature Fusion Network for Drug-Target Interaction Prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Conventional drug discovery is costly and time-consuming.
• Computational drug-target interaction (DTI) prediction offers efficiency and cost reduction.
• Existing DTI methods struggle with effective feature extraction and fusion from protein sequences.

Purpose of the Study:

• To develop an interpretable hybrid deep feature fusion network (IHDFN) for improved DTI prediction.
• To address challenges in combining shallow and deep protein features and enhancing feature fusion complexity.
• To improve drug feature representation and model stability.

Main Methods:

• Hybrid deep feature extraction module for protein sequences using two distinct views.
• StarNet fusion model for efficient shallow and deep feature integration.
• Graph convolutional network (GCN) with residual connections and layer normalization for drug features.
• Attention mechanism for integrating multimodal drug and protein features.

Main Results:

• IHDFN demonstrates exceptional performance and robustness on three datasets.
• The proposed method effectively combines shallow and deep protein features.
• Enhanced feature representation and fusion complexity were achieved.
• Interpretability in DTI prediction was attained through attention mechanisms.

Conclusions:

• IHDFN offers a promising and effective solution for drug-target interaction prediction.
• The model's ability to integrate diverse features and provide interpretability is a significant advancement.
• The findings underscore the potential of IHDFN in accelerating drug discovery pipelines.