ChatMol: interactive molecular discovery with natural language.

Area of Science:

• Biochemistry
• Computational Chemistry
• Artificial Intelligence

Background:

• Natural language processing (NLP) is increasingly vital for human-machine interaction.
• Biochemical tasks like property prediction and molecule mining are complex.
• Integrating natural and chemical languages enhances molecular understanding and task efficiency.

Purpose of the Study:

• Introduce conversational molecular design for describing and editing molecules via natural language.
• Develop ChatMol, a generative pretrained model for this novel task.
• Evaluate ChatMol's effectiveness against existing large language models.

Main Methods:

• Developed ChatMol, a generative pretrained model.
• Enhanced ChatMol with experimental property data and molecular spatial information.
• Incorporated associations between natural and chemical languages.
• Evaluated model performance against benchmarks like ChatGPT.

Main Results:

• Conversational molecular design presents significant challenges.
• ChatMol demonstrates effectiveness, outperforming typical large language models.
• Knowledge enhancement is crucial for successful conversational molecular design.

Conclusions:

• Conversational molecular design is a promising new area for molecular discovery.
• ChatMol provides a strong foundation for natural language-driven molecular manipulation.
• Further research can explore advanced natural language interactions in chemistry.