HKFGCN: A novel multiple kernel fusion framework on graph convolutional network to predict microbe-drug associations

Area of Science:

• Microbiology
• Pharmacology
• Computational Biology

Background:

• Human microbes interact with hosts, regulating drug effectiveness.
• Microbe-drug associations are crucial for drug discovery and antimicrobial development.
• Experimental identification of microbe-drug links is time-consuming and costly.

Purpose of the Study:

• To develop an efficient computational method for predicting microbe-drug associations.
• To reduce the cost and time associated with experimental identification.
• To provide a new computational tool for drug discovery.

Main Methods:

• Developed HKFGCN (Heterogeneous information Kernel Fusion Graph Convolution Network).
• Constructed microbe-drug association networks and similarity networks.
• Extracted topological and Gaussian kernel features from multiple networks.
• Reconstructed bipartite microbe-drug graphs using learned representations.

Main Results:

• HKFGCN demonstrated excellent performance across various datasets.
• The model achieved high accuracy in predicting microbe-drug associations.
• Case studies, including HIV, showed results consistent with existing literature.

Conclusions:

• HKFGCN is a promising computational method for predicting microbe-drug associations.
• The model offers a cost-effective and efficient alternative to experimental methods.
• This approach can accelerate drug discovery and development by identifying novel microbe-drug links.