MGEGFP: a multi-view graph embedding method for gene function prediction based on adaptive estimation with GCN.

Area of Science:

• Computational biology
• Bioinformatics
• Network science

Background:

• Integrating multiple biological networks is crucial for accurate gene function inference.
• Existing methods often fail to fully capture gene neighborhood relationships and adaptively estimate data view contributions.
• Challenges remain in maximizing the benefits of multi-view integration for gene function prediction.

Purpose of the Study:

• To propose MGEGFP, a multi-view graph embedding method leveraging adaptive estimation and Graph Convolutional Networks (GCNs).
• To enhance gene representation learning from multiple interaction networks for improved function prediction.
• To address limitations in previous methods regarding neighborhood relationships and view contribution estimation.

Main Methods:

• Designed a dual-channel GCN encoder to separate view-specific information and consensus patterns.
• Developed a multi-gate module for adaptive estimation of data view contributions during reconstruction.
• Incorporated a diversity preservation constraint to mitigate overfitting in the multi-view integration process.

Main Results:

• MGEGFP demonstrated superior performance in gene function prediction compared to seven state-of-the-art methods on yeast and human datasets.
• Ablation studies confirmed the significant contributions of the dual-channel encoder, multi-gate module, and diversity preservation constraint.
• The method effectively learns high-quality gene representations by leveraging multiplexity advantages.

Conclusions:

• MGEGFP offers a robust and effective approach for gene function prediction through advanced multi-view network integration.
• The proposed adaptive estimation strategy and dual-channel GCN architecture significantly enhance model performance.
• MGEGFP represents a valuable tool for advancing computational approaches in systems biology and functional genomics.