Heterogeneous network drug-target interaction prediction model based on graph wavelet transform and multi-level...

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Accurate drug-target interaction (DTI) prediction is crucial for drug discovery.
• Existing methods face challenges like data imbalance, poor interpretability, and overlooking protein dynamics.

Purpose of the Study:

• To introduce GHCDTI, a heterogeneous graph neural framework to improve DTI prediction.
• To address data imbalance, enhance interpretability, and incorporate protein dynamics.

Main Methods:

• Utilized cross-view contrastive learning with adaptive sampling for imbalanced data.
• Employed heterogeneous data fusion with cross-graph attention for integrated insights.
• Applied multi-scale wavelet feature extraction to capture protein conformational dynamics.

Main Results:

• Achieved state-of-the-art performance on benchmark datasets (AUC: 0.966 ± 0.016; AUPR: 0.888 ± 0.018).
• Demonstrated efficient processing of large datasets (1,512 proteins, 708 drugs in <2 minutes).
• Provided interpretable residue-level insights into drug-target interactions.

Conclusions:

• GHCDTI effectively predicts drug-target pairs, overcoming key limitations in DTI prediction.
• The framework offers a scalable and efficient tool for virtual screening and drug repositioning.
• GHCDTI advances drug discovery and biomedical knowledge integration.