Artificial Intelligence-Guided De Novo Molecular Design Targeting COVID-19

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Traditional methods for identifying SARS-CoV-2 therapeutic agents, like computational docking, face limitations due to high costs and vast chemical spaces.
• Discovering new drug candidates is crucial for combating the ongoing global health crisis.

Purpose of the Study:

• To present a novel de novo molecular design strategy utilizing artificial intelligence (AI) for the discovery of new therapeutic agents against SARS-CoV-2.
• To overcome the limitations of traditional in silico screening methods.

Main Methods:

• Employed an iterative search and retrain strategy combining Monte Carlo tree search with multitask neural network surrogate models and recurrent neural networks.
• Utilized Vina scores to assess binding affinity to the SARS-CoV-2 spike protein and its receptor interface.
• Generated hundreds of new molecular candidates.

Main Results:

• The AI-driven approach successfully generated numerous novel therapeutic agents.
• Identified candidates demonstrated superior binding performance compared to large databases of existing FDA-approved and non-FDA molecules.
• The AI strategy proved broadly applicable for designing molecules with desired functionalities.

Conclusions:

• The developed AI strategy offers a powerful and efficient method for accelerated drug discovery.
• This approach significantly enhances the identification of potential therapeutic agents for SARS-CoV-2 and other diseases.
• The AI methodology is adaptable for discovering chemical molecules with various functional properties.