AntiViralDL: Computational Antiviral Drug Repurposing Using Graph Neural Network and Self-Supervised Learning

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Viral infections pose significant global health challenges.
• Traditional antiviral drug development is resource-intensive and inefficient.
• Computational methods offer a promising alternative for accelerating drug discovery.

Purpose of the Study:

• To develop an efficient computational framework, AntiViralDL, for predicting virus-drug associations.
• To leverage self-supervised learning and graph convolutional networks for enhanced prediction accuracy.
• To address data sparsity in virus-drug association prediction.

Main Methods:

• Constructed a virus-drug association dataset by integrating Drugvirus2 and FDA-approved data.
• Employed Light Graph Convolutional Network (LightGCN) for learning virus and drug embeddings.
• Utilized contrastive learning and data augmentation with random noise for improved prediction.
• Predicted virus-drug associations using an inner product calculation.

Main Results:

• AntiViralDL achieved AUC of 0.8450 and AUPR of 0.8494.
• Outperformed four benchmarked virus-drug association prediction models.
• Demonstrated efficacy in identifying potential anti-COVID-19 drug candidates through a case study.

Conclusions:

• AntiViralDL provides an efficient and accurate computational approach for predicting antiviral drugs.
• The framework effectively addresses data sparsity and improves prediction performance.
• AntiViralDL shows significant potential for accelerating the discovery of novel antiviral therapies.