Generating Potential Protein-Protein Interaction Inhibitor Molecules Based on Physicochemical Properties.

Area of Science:

• Medicinal Chemistry
• Computational Drug Discovery
• Pharmacology

Background:

• Protein-protein interactions (PPIs) are crucial in disease pathogenesis and represent promising drug targets.
• Conventional drug discovery methods and molecular generation models struggle with the unique physicochemical properties of PPI inhibitors, which differ from traditional oral drugs adhering to the Rule of Five (RO5).

Purpose of the Study:

• To develop a specialized molecular generation model for creating novel protein-protein interaction (PPI) inhibitors.
• To address the challenges in designing drugs targeting PPIs due to their distinct physicochemical properties.

Main Methods:

• A deep reinforcement learning (DRL) model was employed for molecular generation.
• A custom scoring function was integrated into the DRL model to represent key properties of PPI inhibitors.

Main Results:

• The DRL model successfully generated potential PPI inhibitor compounds.
• The generated virtual compounds exhibit desired properties characteristic of PPI inhibitors.
• These compounds show structural similarity to existing commercial PPI inhibitor libraries.

Conclusions:

• The proposed DRL-based molecular generation model is effective for discovering PPI inhibitors.
• This approach provides a valuable virtual library of novel PPI inhibitor candidates for further research and development.
• The study offers a new strategy for targeting challenging PPIs in drug discovery.