Identifying potential association on gene-disease network via dual hypergraph regularized least squares

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying gene-disease associations is crucial but experimentally challenging.
• Machine learning models are increasingly used for exploring genetic information in complex diseases.
• Current methods often fail to leverage diverse biological data or capture complex, higher-order relationships.

Purpose of the Study:

• To develop a novel computational method for detecting potential gene-disease associations.
• To overcome limitations of existing methods by integrating multiple biological data sources.
• To capture higher-order relationships among genes and diseases for improved prediction accuracy.

Main Methods:

• Proposed Dual Hypergraph Regularized Least Squares (DHRLS) method.
• Utilized Centered Kernel Alignment-based Multiple Kernel Learning (CKA-MKL) to optimize kernels from various biological data.
• Employed hypergraphs to model higher-order relationships and Alternating Least Squares (ALSA) for model solving.

Main Results:

• DHRLS demonstrated superior performance in predicting gene-disease associations compared to existing tools.
• The method achieved excellent prediction accuracy across six real-world biological networks under cross-validation.
• Validated the robustness and effectiveness of the proposed DHRLS approach.

Conclusions:

• DHRLS significantly enhances the discovery of potential disease-associated genes.
• The approach provides valuable guidance for experimental verification in complex disease research.
• Highlights the potential of hypergraph-based methods and multiple kernel learning in bioinformatics.