A dual-scale fused hypergraph convolution-based hyperedge prediction model for predicting missing reactions in...

Area of Science:

• Systems Biology
• Computational Biology
• Metabolic Engineering

Background:

• Genome-scale metabolic models (GEMs) are crucial for understanding cellular metabolism but often contain incomplete reaction data.
• Existing methods for filling gaps in GEMs lack the ability to differentiate substrates from products, limiting predictive accuracy.

Purpose of the Study:

• To develop an advanced method for inferring missing reactions in GEMs by effectively distinguishing substrates and products.
• To enhance the predictive performance of GEMs through improved gap-filling strategies.

Main Methods:

• Proposed DSHCNet, a hyperedge prediction model utilizing dual-scale fused hypergraph convolution.
• Modeled hyperedges as heterogeneous complete graphs, decomposed them into homogeneous and heterogeneous subgraphs.
• Employed graph convolution and attention mechanisms to extract and fuse vertex features, distinguishing substrate and product information.

Main Results:

• DSHCNet achieved an average recovery rate of missing reactions at least 11.7% higher than state-of-the-art methods.
• GEMs reconstructed using DSHCNet demonstrated superior predictive performance.
• The dual-scale graph decomposition enhanced information propagation and feature distinguishability.

Conclusions:

• DSHCNet effectively addresses the limitations of previous gap-filling methods by differentiating substrates and products.
• The proposed model significantly improves the accuracy and predictive power of genome-scale metabolic models.
• DSHCNet offers a superior approach for completing metabolic networks in systems biology research.