Pocket-Based Generative Diffusion Model Accelerates Potent Influenza A Hemagglutinin Inhibitor Discovery

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Deep generative models advance 3D chemical exploration but often neglect target affinity and structural rationality.
• Balancing binding affinity with molecular structure is crucial for effective drug development.

Purpose of the Study:

• To develop a novel dual conditional diffusion model (DCDM) for refined 3D target-based molecular generation.
• To enhance predicted binding affinity and maintain structural rationality and diversity in generated molecules.

Main Methods:

• Established a dual conditional diffusion model (DCDM) incorporating ligand-protein interaction features.
• Applied DCDM to optimize penindolone (PND) targeting influenza A hemagglutinin (HA).
• Synthesized and evaluated DCDM-inspired derivatives.

Main Results:

• DCDM demonstrated superiority in enhancing binding affinity and maintaining structural rationality and diversity.
• A promising candidate, compound C2e, showed a 26-fold higher affinity for HA and a 10-fold lower IC50 than PND.
• Compound C2e exhibited potent in vivo antiviral activity and favorable safety profiles.

Conclusions:

• DCDM is a valuable generative model for accelerating drug development.
• The model effectively balances target affinity and structural rationality in molecular generation.
• Optimized compounds show significant therapeutic potential for influenza treatment.