Inferring Metabolite-Disease Association Using Graph Convolutional Networks.

Area of Science:

• Computational biology
• Bioinformatics
• Network science

Background:

• Biological experiments are time-consuming and resource-intensive.
• Existing computational models often overlook topological structures in metabolite-disease association networks.
• Developing efficient computational tools is crucial for advancing biological research.

Purpose of the Study:

• To propose a novel graph convolutional network-based method (MDAGCN) for inferring metabolite-disease associations.
• To address the limitations of current methods in capturing network topology.
• To improve the accuracy and reliability of predicting potential metabolite-disease relationships.

Main Methods:

• Calculated and integrated multiple metabolite and disease similarity metrics, filtering noise.
• Constructed a heterogeneous network integrating metabolite similarity, disease similarity, and known associations.
• Applied graph convolutional networks (GCNs) to learn node features and infer potential associations.

Main Results:

• MDAGCN demonstrated superior performance in cross-validation compared to existing methods.
• Case studies validated the reliability and effectiveness of the proposed MDAGCN model.
• The method successfully inferred potential metabolite-disease associations by leveraging network topology.

Conclusions:

• MDAGCN offers a more reliable approach for predicting metabolite-disease associations.
• The integration of network topology analysis enhances the predictive power of computational models.
• This GCN-based method provides a valuable tool for biological research and drug discovery.