COVID-19 Multi-Targeted Drug Repurposing Using Few-Shot Learning.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Pharmacology

Background:

• COVID-19 treatment necessitates multi-targeted therapies, but computational screening is hindered by limited data.
• Existing methods struggle with scarce experimental data and complex molecular information extraction.

Purpose of the Study:

• Introduce MolGNN, a deep learning model for molecular property prediction.
• Develop a robust few-shot learning tool for identifying potential COVID-19 therapeutics.
• Facilitate multi-targeted drug discovery for complex diseases like COVID-19.

Main Methods:

• Utilized a graph neural network (GNN) for molecular embedding and property prediction.
• Employed a deep learning approach enabling effective learning without extensive manual labels.
• Validated performance on datasets with limited labeled experimental data.

Main Results:

• MolGNN achieved comparable or superior prediction performance to state-of-the-art methods, even with minimal labeled data.
• Demonstrated robustness in few-shot learning scenarios, proving effective with scarce training data.
• Identified multi-targeted molecules for Janus kinases (cytokine storm) and SARS-CoV-2 main protease (viral replication).

Conclusions:

• MolGNN is a powerful few-shot learning tool for molecular property prediction, adept at handling limited data.
• The model successfully predicted promising drug candidates for COVID-19, targeting key viral and host factors.
• Predicted molecules show potential for treating COVID-19 symptoms and inhibiting viral replication, with some supported by existing evidence.