Association Mining to Identify Microbe Drug Interactions Based on Heterogeneous Network Embedding Representation.

Area of Science:

• Computational biology
• Bioinformatics
• Pharmacology

Background:

• Accurate microbe-drug association identification is crucial for drug development and precision medicine.
• Wet-lab methods are time-consuming and expensive, necessitating computational approaches.
• Limited computational tools exist for microbe-drug association prediction.

Purpose of the Study:

• To develop a novel computational framework for predicting microbe-drug associations.
• To leverage heterogeneous biomedical data for improved prediction accuracy.
• To address the limitations of existing methods in microbe-drug association prediction.

Main Methods:

• Constructed a heterogeneous network integrating drug-drug, microbe-microbe, and microbe-drug interactions.
• Proposed the Heterogeneous Network Embedding Representation for Microbe-Drug Association (HNERMDA) framework.
• Employed metapath2vec for learning low-dimensional microbe and drug embeddings.
• Utilized a bias bipartite network projection recommendation algorithm for enhanced prediction.

Main Results:

• HNERMDA consistently outperformed five baseline methods across three cross-validation types on MDAD and aBiofilm datasets.
• The model demonstrated effectiveness in inferring potential target microbes for drugs through case studies.
• Achieved superior prediction accuracy compared to existing computational methods.

Conclusions:

• HNERMDA provides an effective computational approach for predicting microbe-drug associations.
• The framework enhances drug development and precision medicine by identifying microbial targets.
• This method offers a valuable alternative to traditional wet-lab techniques.