Attending on Multilevel Structure of Proteins enables Accurate Prediction of Cold-Start Drug-Target Interactions

2025-10-05

AI Summary

Existing cold-start drug–target interaction (DTI) prediction methods rely solely on protein primary structure, limiting their representational capacity and biological interpretability. Method: We propose a hierarchical attention model that systematically integrates protein structural features across all four levels (primary to quaternary). A hierarchical attention mechanism captures both local and global interactions between drugs and each structural level, while multi-granularity feature fusion and structure-aware representation learning jointly encode drug–protein pairs. Contribution/Results: The model significantly enhances biological interpretability and generalization capability. Extensive experiments on multiple benchmark datasets demonstrate that our approach achieves substantial improvements in AUC and AUPR over state-of-the-art methods under cold-start settings, empirically validating the critical value of multi-level structural information for DTI prediction.

Technology Category

• Machine Learning: Structured Learning
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Cold-start drug-target interaction (DTI) prediction focuses on interaction between novel drugs and proteins. Previous methods typically learn transferable interaction patterns between structures of drug and proteins to tackle it. However, insight from proteomics suggest that protein have multi-level structures and they all influence the DTI. Existing works usually represent protein with only primary structures, limiting their ability to capture interactions involving higher-level structures. Inspired by this insight, we propose ColdDTI, a framework attending on protein multi-level structure for cold-start DTI prediction. We employ hierarchical attention mechanism to mine interaction between multi-level protein structures (from primary to quaternary) and drug structures at both local and global granularities. Then, we leverage mined interactions to fuse structure representations of different levels for final prediction. Our design captures biologically transferable priors, avoiding the risk of overfitting caused by excessive reliance on representation learning. Experiments on benchmark datasets demonstrate that ColdDTI consistently outperforms previous methods in cold-start settings.

Problem

• Predicting interactions between novel drugs and unknown proteins
• Capturing multi-level protein structures from primary to quaternary
• Addressing limitations of existing single-structure representation methods

Innovation

• Hierarchical attention mechanism for multi-level protein structures
• Interaction mining between protein and drug structures
• Fusing structure representations for final prediction