Graph-based drug-target interaction modeling: from representation learning to output-driven drug discovery

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Graph-based deep learning is a powerful framework for modeling drug-target interactions (DTIs).
• These models integrate molecular, structural, and systems-level information into a unified representation.
• Existing models span network-, sequence/hybrid-, and structure-based paradigms.

Purpose of the Study:

• To survey graph-based DTI models across different paradigms.
• To introduce an output-driven evaluation perspective aligned with drug discovery stages.
• To discuss advancements, applications, challenges, and future directions.

Main Methods:

• Review of graph-based DTI models including network-, sequence/hybrid-, and structure-based approaches.
• Evaluation framework based on model output alignment with drug discovery pipeline needs.
• Analysis of attention mechanisms, semi-supervised learning, and benchmarking challenges.

Main Results:

• Graph-based DTI models offer a continuum from association inference to structure-resolved interaction modeling.
• Attention mechanisms and semi-supervised learning enhance data efficiency and feature prioritization.
• Model outputs support target identification, drug repurposing, and lead optimization.

Conclusions:

• A unified perspective on the design, evaluation, and translational application of graph-based DTI models is provided.
• Addressing benchmarking challenges like data leakage and structural bias is crucial.
• Emerging directions include multimodal integration and the use of predicted protein structures.