CReM: chemically reasonable mutations framework for structure generation.

Area of Science:

• Computational chemistry
• Drug discovery
• cheminformatics

Background:

• De novo molecular design is crucial for discovering novel drug candidates.
• Current structure generation methods, including deep learning and atom-based approaches, often produce invalid or synthetically challenging molecules.
• Reaction-based and fragment-based methods have limitations in novelty, diversity, or addressing synthetic complexity.

Purpose of the Study:

• To develop a novel fragment-based framework for de novo structure generation.
• To ensure chemically valid structures with controllable synthetic feasibility.
• To enhance novelty, diversity, and control over chemotypes in generated compounds.

Main Methods:

• A new fragment-based framework for structure generation was developed.
• The framework inherently produces chemically valid structures.
• Flexible control over diversity, novelty, synthetic complexity, and chemotypes is integrated.

Main Results:

• The developed framework generates chemically valid molecular structures.
• It offers tunable control over key design parameters like novelty and synthetic complexity.
• The approach enhances both diversity and novelty compared to existing methods.

Conclusions:

• The new fragment-based framework addresses limitations of previous de novo design tools.
• It provides a versatile and controllable method for exploring chemical space.
• The open-source Python module facilitates custom workflow creation for researchers.