Neighborhood Topology-Aware Knowledge Graph Learning and Microbial Preference Inferring for Drug-Microbe Association...

Area of Science:

• Pharmacogenomics and Computational Biology
• Microbiome Research and Drug Interactions

Background:

• The human microbiota significantly influences drug efficacy by metabolizing pharmacological agents.
• Existing computational methods for predicting microbe-drug associations often overlook complex network structures and entity-specific preferences.

Purpose of the Study:

• To develop a novel computational method, PCMDA, for accurately predicting microbe-drug associations.
• To address limitations in previous methods by incorporating neighborhood topologies, association preferences, and multi-perspective features.

Main Methods:

• Construction of a comprehensive microbe-disease-drug knowledge graph.
• Generation of topological embeddings using random walks with neighborhood restarts and distance-level attention.
• Integration of topological structure and relational semantics via a knowledge graph learning module and multilayer perceptron networks.
• Incorporation of microbial drug preference using information-level attention and a dual-gated network for feature encoding.

Main Results:

• PCMDA demonstrated superior performance in microbe-drug association prediction compared to seven state-of-the-art methods.
• Case studies confirmed PCMDA's ability to identify reliable candidate microbes associated with specific drugs.

Conclusions:

• PCMDA effectively leverages topological information and entity-specific preferences for accurate microbe-drug association prediction.
• The developed method offers a valuable tool for understanding host-microbe-drug interactions and optimizing drug therapy.