Multiview Deep Learning-Based Molecule Design and Structural Optimization Accelerates Inhibitor Discover.

Area of Science:

• Computational chemistry and drug discovery
• Artificial intelligence in molecular modeling
• Deep generative models for de novo design

Background:

• Developing effective SARS-CoV-2 inhibitors is crucial for combating the COVID-19 pandemic.
• Existing molecular generation models face challenges in producing valid, novel, and property-optimized molecules.
• The need for advanced computational tools to accelerate drug discovery and design.

Purpose of the Study:

• To introduce MEDICO, a multiview deep generative model for molecule generation and optimization.
• To demonstrate MEDICO's capability in discovering SARS-CoV-2 inhibitors.
• To enhance the generation of molecules with desired structural and chemical properties.

Main Methods:

• Utilizing a multiview representation learning framework for comprehensive structural semantics.
• Employing graph generative models for molecular graph generation.
• Integrating molecular docking as a chemical prior for targeted drug design.
• Applying MEDICO to structural optimization of known SARS-CoV-2 inhibitors.

Main Results:

• MEDICO significantly outperforms state-of-the-art methods in generating valid, novel, and unique molecules, with an 85% improvement in validity.
• The model successfully generates molecules with desired drug-like properties, structurally similar to targeted molecules.
• Case studies show successful generation of novel Mpro inhibitors for SARS-CoV-2.
• Structural optimization of known inhibitors resulted in an 88% improvement in binding affinity to Mpro.

Conclusions:

• MEDICO represents a significant advancement in AI-driven molecular generation and optimization.
• The model's multiview approach enhances the learning of molecular topology and geometry.
• MEDICO shows strong potential for accelerating the de novo design of therapeutics for SARS-CoV-2 and other diseases.