HyperDC: A Non-Uniform Hypergraph Framework for Dual- and Higher-Order Drug Combination Recommendation Across Diverse...

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Complex diseases involve multiple targets, necessitating combination therapies.
• Current computational methods for drug combination prediction face challenges like cross-disease application and handling low-resource diseases.

Purpose of the Study:

• To develop a unified, cross-disease framework for recommending drug combinations.
• To address limitations of existing methods in predicting dual- and multidrug combinations and handling data scarcity.

Main Methods:

• Constructed a nonuniform hypergraph using clinical and knowledge-driven drug-disease association data.
• Integrated knowledge graph pretraining and adversarial negative sampling for enhanced model discrimination.
• Unified modeling space for single drugs and dual/multidrug combinations.

Main Results:

• HyperDC outperformed existing methods in unified dual-drug benchmarks and disease-specific tasks.
• Demonstrated superior performance in clinical ranking tasks, identifying FDA-approved and clinically supported combinations.
• Successfully applied to a data-sparse MASH scenario, with 66.7% of tested combinations showing synergistic effects.

Conclusions:

• HyperDC offers a unified approach for cross-disease drug combination recommendation.
• The framework improves prioritization efficiency and narrows screening space for combination therapies.
• Provides methodological support for developing combination strategies for complex diseases, validated by *in vitro* experiments.