A geometric deep learning model for display and prediction of potential drug-virus interactions against SARS-CoV-2

Area of Science:

• Computational Biology
• Artificial Intelligence in Medicine
• Drug Discovery

Background:

• The COVID-19 pandemic, despite reduced lethality due to vaccines and immunity, poses ongoing risks from new variants and animal reservoirs.
• Predicting future viral threats and developing rapid countermeasures requires efficient methods for examining drug-virus interactions.
• Traditional experimental methods for drug-virus interaction analysis are costly, time-consuming, and labor-intensive.

Purpose of the Study:

• To propose a novel geometric deep learning (GDL) model for predicting drug-virus interactions against COVID-19.
• To leverage GDL as a faster and more cost-effective alternative to experimental approaches.
• To enhance the preparedness against potential future viral variants and pandemics.

Main Methods:

• Utilized antiviral drug data in Simplified Molecular Input Line Entry System (SMILES) format to generate descriptive molecular features.
• Converted molecular data into a graph structure suitable for GDL models.
• Employed a Message Passing Neural Network (MPNN) for node feature vector transformation and training.
• Developed a geometric neural network architecture with fully connected layers for prediction.

Main Results:

• The proposed GDL model demonstrated high performance in predicting drug-virus interactions.
• Achieved an accuracy of 97% in predicting drug-virus interactions, outperforming existing methods.
• Successfully converted complex molecular data into a format processable by GDL.

Conclusions:

• Geometric deep learning presents a viable and efficient alternative for predicting drug-virus interactions.
• The developed GDL model shows significant promise for accelerating drug discovery and vaccine development.
• This approach can aid in rapid response to emerging viral threats and future pandemics.