Reinvent 4: Modern AI-driven generative molecule design

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Machine learning for molecular design

Background:

• Generative artificial intelligence (AI) is revolutionizing molecular design.
• Existing tools often lack comprehensive, open-source implementations of common AI algorithms.
• Drug discovery necessitates efficient and innovative molecular design strategies.

Purpose of the Study:

• To introduce REINVENT 4, an open-source generative AI framework for small molecule design.
• To provide reference implementations of key AI algorithms for molecular generation.
• To foster education, collaboration, and innovation in AI-driven molecular design.

Main Methods:

• Utilizes recurrent neural networks and transformer architectures for molecule generation.
• Integrates generators with machine learning optimization, transfer learning, reinforcement learning, and curriculum learning.
• Operates as a command-line tool with TOML or JSON configuration.

Main Results:

• REINVENT 4 enables de novo design, R-group replacement, library design, linker design, scaffold hopping, and molecule optimization.
• The framework serves as a production tool supporting in-house drug discovery projects.
• Provides a unified, documented codebase for common AI algorithms in molecular design.

Conclusions:

• REINVENT 4 offers a transparent and accessible platform for AI-based molecular design.
• The open-source release promotes wider adoption, collaboration, and advancement in the field.
• Facilitates both research and practical application in drug discovery and chemical innovation.