CF-DTI: coarse-to-fine feature extraction for enhanced drug-target interaction prediction

Area of Science:

• Computational biology
• Drug discovery and development
• Bioinformatics

Background:

• Accurate drug-target interaction (DTI) prediction is vital for efficient drug development.
• Existing DTI prediction models often underutilize multi-level interaction features, limiting performance.
• There is a need for advanced models that capture complex drug-target relationships more effectively.

Purpose of the Study:

• To develop a novel coarse-to-fine drug-target interaction model (CF-DTI).
• To enhance DTI prediction accuracy by integrating coarse-grained and fine-grained features.
• To improve data utilization and model performance in drug discovery.

Main Methods:

• Proposed CF-DTI model integrating coarse- and fine-grained features.
• Implemented an information filtering module with focal and cross-attention mechanisms for feature extraction.
• Employed multi-granularity learning and adaptive fusion strategies for feature integration.

Main Results:

• CF-DTI consistently outperformed existing baseline models across four benchmark datasets.
• Achieved an average improvement of ~2% over the second-best models on large-scale datasets (BindingDB, BioSNAP).
• Demonstrated significant gains in AUROC and AUPRC, particularly in the unseen-pair setting (6% AUROC on BioSNAP).

Conclusions:

• CF-DTI effectively leverages multi-level features for superior DTI prediction accuracy.
• The model shows potential for reducing prediction ambiguity and capturing complex molecular interactions.
• CF-DTI offers a promising tool to accelerate and enhance the drug discovery process.