Scaffold Hopping with Generative Reinforcement Learning.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Scaffold hopping is crucial for drug discovery but challenging for chemists and computational methods.
• Existing methods often limit scaffold hopping to predefined substructures, restricting chemical space exploration.
• Generative reinforcement learning offers potential for accelerating scaffold hopping through property optimization.

Purpose of the Study:

• To advance reinforcement learning for scaffold hopping by enabling unconstrained, full-molecule generation.
• To develop a method that generates novel scaffolds with high 3D and pharmacophore similarity but low scaffold similarity to a reference molecule.
• To demonstrate the effectiveness of the proposed approach in exploring analogs and designing new scaffold-hopping candidates.

Main Methods:

• Introduced the Reinforcement Learning for Unconstrained Scaffold Hopping (RuSH) approach.
• RuSH utilizes generative reinforcement learning for full-molecule generation.
• The method optimizes for high 3D and pharmacophore similarity while minimizing scaffold similarity.

Main Results:

• RuSH demonstrated flexibility and effectiveness in exploring analogs of known scaffold-hops.
• The approach successfully designed scaffold-hopping candidates matching known binding mechanisms.
• Comparison with established methods highlighted RuSH's ability to achieve scaffold diversity and preserve 3D properties.

Conclusions:

• RuSH advances scaffold hopping by enabling unconstrained, full-molecule generation.
• The method systematically achieves scaffold diversity while maintaining critical three-dimensional properties.
• RuSH offers a powerful tool for accelerating drug discovery through novel scaffold design.