Predicting human microbe-drug associations via graph convolutional network with conditional random field.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Pharmacogenomics

Background:

• Human microbes are crucial for drug development and precision medicine.
• Understanding microbe-drug interactions is complex and challenging.
• Computational approaches offer a cost-effective alternative to experimental methods.

Purpose of the Study:

• To develop a computational framework for predicting human microbe-drug associations.
• To leverage biological information and network construction for improved predictions.
• To address the limitations of existing computational methods in this field.

Main Methods:

• Constructed a heterogeneous network including microbe and drug similarity networks, and a microbe-drug interaction network.
• Proposed a novel graph convolutional network (GCN)-based framework (GCNMDA).
• Integrated Conditional Random Field (CRF) with an attention mechanism and employed a random walk with restart for feature learning.

Main Results:

• GCNMDA demonstrated superior performance compared to seven state-of-the-art methods across three datasets.
• The model successfully identified potential microbe-drug associations in case studies, including SARS-CoV-2, Ciprofloxacin, and Moxifloxacin.
• Experimental results validated the effectiveness of the proposed GCNMDA framework.

Conclusions:

• GCNMDA provides an effective computational approach for predicting microbe-drug associations.
• The framework enhances insights into microbe-drug interaction mechanisms.
• This work facilitates advancements in drug discovery, repurposing, and precision medicine.