MDGNN: Microbial Drug Prediction Based on Heterogeneous Multi-Attention Graph Neural Network

Area of Science:

• Computational biology
• Drug discovery
• Machine learning

Background:

• The COVID-19 pandemic highlights the urgent need for rapid antiviral drug discovery.
• Traditional methods struggle to keep pace with emerging infectious diseases.
• Predicting drug-virus associations is crucial for understanding interactions and screening antivirals.

Purpose of the Study:

• To develop a deep learning algorithm for predicting potential antiviral drugs.
• To enhance the prediction of drug-virus associations.
• To guide the screening process for effective antiviral therapies.

Main Methods:

• Developed a novel deep learning algorithm named MDGNN (Multi-dimensional Graph Neural Network).
• MDGNN incorporates node-level and feature-level attention mechanisms within graph convolutions.
• Integrated global graph information during the information aggregation process.

Main Results:

• MDGNN demonstrated state-of-the-art performance in predicting drug-virus associations.
• Achieved an Area Under the Curve (AUC) of 0.9726 and Area Under the Precision-Recall Curve (AUPR) of 0.9112.
• Successfully predicted and validated two drugs relevant to SARS-CoV-2 through literature review.

Conclusions:

• MDGNN is a highly effective deep learning model for predicting antiviral drugs.
• The model's attention mechanisms improve the integration of graph information for better predictions.
• MDGNN offers a valuable tool for accelerating antiviral drug discovery and combating infectious diseases.